Spatio-temporal scene graphs represent interactions in a video by decomposing scenes into individual objects and their pair-wise temporal relationships. Long-term anticipation of the fine-grained pair-wise relationships between objects is a challenging problem. To this end, we introduce the task of Scene Graph Anticipation (SGA). We adopt state-of-the-art scene graph generation methods as baselines to anticipate future pair-wise relationships between objects. In our proposed approaches, we leverage object-centric representations of relationships to reason about the observed video frames and model the evolution of relationships between objects. We take a continuous time perspective and model the latent dynamics of the evolution of object interactions using concepts of NeuralODE and NeuralSDE, respectively. We infer representations of future relationships by solving an Ordinary Differential Equation and a Stochastic Differential Equation, respectively. Extensive experimentation on the Action Genome dataset validates the efficacy of the proposed methods.

Category-level articulated object pose estimation task focuses on the pose estimation of unknown articulated objects within known categories. Despite its significance, this task remains challenging due to the varying shapes and poses of objects, expensive dataset annotation costs, and complex real-world environments. In this paper, we propose a novel self-supervised approach that leverages a single-frame point cloud to accomplish the aforementioned task. Our model consistently generates canonical reconstruction with a canonical pose and joint state for the entire input object, and it estimates object-level pose that reduce overall pose variance and part-level poses that align each part of the input with its corresponding part of the reconstruction. Experimental results demonstrate our approach achieves state-of-the-art performance to other self-supervised methods and comparable performance to other supervised methods. To assess the performance of our model in real-world scenarios, we also introduce a novel real-world articulated object benchmark dataset.

Computer vision methods that explicitly detect object parts and reason on them are a step towards inherently interpretable models. Existing approaches that perform part discovery driven by a fine-grained classification task make very restrictive assumptions on the geometric properties of the discovered parts; they should be small and compact. Although this prior is useful in some cases, in this paper we show that pre-trained transformer-based vision models, such as self-supervised DINOv2 ViT, enable the relaxation of these constraints. In particular, we find that a total variation (TV) prior, which allows for multiple connected components of any size, substantially outperforms previous work. We test our approach on three fine-grained classification benchmarks: CUB, PartImageNet and Oxford Flowers, and compare our results to previously published methods as well as a re-implementation of the state-of-the-art method PDiscoNet with a transformer-based backbone. We consistently obtain substantial improvements across the board, both on part discovery metrics and the downstream classification task, showing that the strong inductive biases in self-supervised ViT models require to rethink the geometric priors that can be used for unsupervised part discovery.

3D dense captioning is a task to localize objects and generate descriptions for each object in a 3D scene. Recent approaches in 3D dense captioning have attempted to incorporate contextual information by modeling relationships with object pairs or aggregating nearest neighbor features of an object. However, the contextual information constructed in these scenarios is limited in two aspects: first, objects have multiple positional relationship that exists across the entire global scene (not only near the object itself), and second, it faces with contradicting objectives--where localization and attribute descriptions are generated better with tight localization, while descriptions involving global positional relations are generated better with contextualized features of the global scene. To overcome this challenge, we introduce CSI, a transformer encoder-decoder pipeline that engages in 3D dense captioning for each object with contextualized geometries (where the structural contexts relevant to each object is summarized) and contextualized objects (where the objects relevant to the summarized structural contexts are aggregated). This simple extension relieves previous methods from the contradicting objectives, enhancing both localization performance while enabling to aggregate contextual features throughout the global scene; thus improving caption generation performance simultaneously. Extensive experiments on two of the most widely-used 3D dense captioning datasets (ScanRefer …

We propose OmniNOCS: a large-scale monocular dataset with 3D Normalized Object Coordinate Space (NOCS), object masks, and 3D bounding box annotations for indoor and outdoor scenes. OmniNOCS has 20 times more object classes and 200 times more instances than existing NOCs datasets (NOCS-Real275, Wild6D). We use OmniNOCS to train a novel, transformer-based monocular NOCS prediction model (NOCSformer) that can predict accurate NOCS, instance masks and poses from 2D object detections across diverse classes. It is the first NOCS model that can generalize to a broad range of classes when prompted with 2D boxes. We evaluate our model on the task of 3D oriented bounding box prediction, where it achieves comparable results to state-of-the-art 3D detection methods such as Cube R-CNN. Unlike other 3D detection methods, our model also provides detailed and accurate 3D object shape and segmentation. We propose a novel benchmark for the task of NOCS prediction based on OmniNOCS, which we hope will serve as a useful baseline for future work in this area. Our dataset and code are at the project website: https://omninocs.github.io

Class-agnostic counting methods enumerate objects of an arbitrary class, providing tremendous utility in many fields. Prior works have limited usefulness as they require either a set of examples of the type to be counted or that the query image contains only a single type of object. A significant factor in these shortcomings is the lack of a dataset to properly address counting in settings with more than one kind of object present. To address these issues, we propose the first Multi-class, Class-Agnostic Counting dataset (MCAC) and A Blind Counter (ABC123), a method that can count multiple types of objects simultaneously without using examples of type during training or inference. ABC123 introduces a new paradigm where instead of requiring exemplars to guide the enumeration, examples are found after the counting stage to help a user understand the generated outputs. We show that ABC123 outperforms contemporary methods on MCAC without needing human in-the-loop annotations. We also show that this performance transfers to FSC-147, the standard class-agnostic counting dataset. MCAC is available at MCAC.active.vision and ABC123 is available at ABC123.active.vision.

In panoptic scene graph generation (PSGG), models retrieve interactions between objects in an image which are grounded by panoptic segmentation masks. Previous evaluations on panoptic scene graphs have been subject to an erroneous evaluation protocol where multiple masks for the same object can lead to multiple relation distributions per mask-mask pair. This can be exploited to increase the final score. We correct this flaw and provide a fair ranking over a wide range of existing PSGG models. The observed scores for existing methods increase by up to 7.4 mR@50 for all two-stage methods, while dropping by up to 19.3 mR@50 for all one-stage methods, highlighting the importance of a correct evaluation. Contrary to recent publications, we show that existing two-stage methods are competitive to one-stage methods. Building on this, we introduce the Decoupled SceneFormer (DSFormer), a novel two-stage model that outperforms all existing scene graph models by a large margin of +11 mR@50 and +10 mNgR@50 on the corrected evaluation, thus setting a new SOTA. As a core design principle, DSFormer encodes subject and object masks directly into feature space.

Scene Graph Generation (SGG) offers a structured representation critical in many computer vision applications.Traditional SGG approaches, however, are limited by a closed-set assumption, restricting their ability to recognize only predefined object and relation categories. To overcome this, we categorize SGG scenarios into four distinct settings based on the node and edge: Closed-set SGG, Open Vocabulary (object) Detection-based SGG (OvD-SGG), Open Vocabulary Relation-based SGG (OvR-SGG), and Open Vocabulary Detection + Relation-based SGG (OvD+R-SGG). While object-centric open vocabulary SGG has been studied recently, the more challenging problem of relation-involved open-vocabulary SGG remains relatively unexplored. To fill this gap, we propose a unified framework named OvSGTR towards fully open vocabulary SGG from a holistic view. The proposed framework is an end-to-end transformer architecture, which learns a visual-concept alignment for both nodes and edges, enabling the model to recognize unseen categories. For the more challenging settings of relation-involved open vocabulary SGG, the proposed approach integrates relation-aware pre-training utilizing image-caption data and retains visual-concept alignment through knowledge distillation. Comprehensive experimental results on the Visual Genome benchmark demonstrate the effectiveness and superiority of the proposed framework.

Brain-inspired Spiking Neural Networks (SNNs) have bio-plausibility and low-power advantages over Artificial Neural Networks (ANNs). Applications of SNNs are currently limited to simple classification tasks because of their poor performance. In this work, we focus on bridging the performance gap between ANNs and SNNs on object detection. Our design revolves around network architecture and spiking neuron. First, the overly complex module design causes spike degradation when the YOLO series is converted to the corresponding spiking version. We design a SpikeYOLO architecture to solve this problem by simplifying the vanilla YOLO and incorporating meta SNN blocks. Second, object detection is more sensitive to quantization errors in the conversion of membrane potentials into binary spikes by spiking neurons. To address this challenge, we design a new spiking neuron that activates Integer values during training while maintaining spike-driven by extending virtual timestep during inference. The proposed method is validated on both static and neuromorphic object detection datasets. On the static COCO dataset, we obtain 66.2% mAP@50 and 48.9% mAP@50:95, which is +15.0% and +18.7% higher than the prior state-of-the-art SNN, respectively. On the neuromorphic Gen1 dataset, we achieve 67.2% mAP@50, which is +8.2% and +2.5% greater than the existing best SNN model and …

Snapshot compressive spectral imaging reconstruction aims to reconstruct three-dimensional spatial-spectral images from a single-shot two-dimensional compressed measurement. Existing state-of-the-art methods are mostly based on deep unfolding structures but have intrinsic performance bottlenecks: i) the ill-posed problem of dealing with heavily degraded measurement, and ii) the regression loss-based reconstruction models being prone to recover images with few details. In this paper, we introduce a generative model, namely the latent diffusion model (LDM), to generate degradation-free prior to enhance the regression-based deep unfolding method by a two-stage training procedure. Furthermore, we propose a Trident Transformer (TT), which extracts correlations among prior knowledge, spatial, and spectral features, to integrate knowledge priors in deep unfolding denoiser, and guide the reconstruction for compensating high-quality spectral signal details. To our knowledge, this is the first approach to integrate physics-driven deep unfolding with generative LDM in the context of CASSI reconstruction. Numeric and visual comparisons on synthetic and real-world datasets illustrate the superiority of our proposed method in both reconstruction quality and computational efficiency. Code will be released.

We introduce RAFT2, a faster, simpler, and more accurate RAFT for optical flow. Compared with RAFT, RAFT2 is supervised with a mixture of Laplace loss. It directly regresses an initial flow for faster convergence in recurrent refinements and introduces stereo pretraining to improve generalization. RAFT2 achieves state-of-the-art on Spring benchmark with 3.69 end-point-error (EPE) and 0.36 1-pixel outlier rate (1px), representing 22.9% and 17.8% error reduction from best-published results. In addition, RAFT2 obtains the best cross-dataset generalization on KITTI(train) and Spring(train). With its high efficiency, RAFT2 operates at least 2.3x faster than mainstream methods while maintaining competitive performance, advancing the state of recurrent refinement frameworks in optical flow estimation.

Single-photon cameras (SPCs) are emerging as sensors of choice for various challenging imaging applications. One class of SPCs based on the single-photon avalanche diode (SPAD) detects individual photons using an avalanche process; the raw photon data can then be processed to extract scene information under extremely low light, high dynamic range, and rapid motion. Yet, single-photon sensitivity in SPADs comes at a cost --- each photon detection consumes more energy than that of a CMOS camera. This avalanche power significantly limits sensor resolution and could restrict widespread adoption of SPAD-based SPCs. We propose a computational-imaging approach called photon inhibition to address this challenge. Photon inhibition strategically allocates detections in space and time based on downstream inference task goals and resource constraints. We develop lightweight, on-sensor computational inhibition policies that use past photon data to disable SPAD pixels in real-time, to select the most informative future photons. As case studies, we design policies tailored for image reconstruction and edge detection, and demonstrate, both via simulations and real SPC captured data, considerable reduction in photon detections (over 90% of photons) while maintaining task performance metrics. Our work raises the question of ``which photons should be detected?'', and paves the way for future …

A minimalist vision system uses the smallest number of pixels needed to solve a vision task. While traditional cameras use a large grid of square pixels, a minimalist camera uses freeform pixels that can take on arbitrary shapes to increase their information content. We show that the hardware of a minimalist camera can be modeled as the first layer of a neural network, where the subsequent layers are used for inference. Training the network for any given task yields the shapes of the camera's freeform pixels, each of which is implemented using a photodetector and an optical mask. We have designed minimalist cameras for monitoring indoor spaces (with 8 pixels), measuring room lighting (with 8 pixels), and estimating traffic flow (with 8 pixels). The performance demonstrated by these systems is on par with a traditional camera with orders of magnitude more pixels. Minimalist vision has two major advantages. First, it naturally tends to preserve the privacy of individuals in the scene since the captured information is inadequate for extracting visual details. Second, since the number of measurements made by a minimalist camera is very small, we show that it can be fully self-powered, i.e., function without an external power supply …

We present an imaging and neural rendering technique that seeks to synthesize videos of light propagating through a scene from novel, moving camera viewpoints. Our approach relies on a new ultrafast imaging setup to capture a first-of-its kind, multi-viewpoint video dataset with picosecond-level temporal resolution. Combined with this dataset, we introduce an efficient neural volume rendering framework based on the transient field. This field is defined as a mapping from a 3D point and 2D direction to a high-dimensional, discrete-time signal that represents time-varying radiance at ultrafast timescales. Rendering with transient fields naturally accounts for effects due to the finite speed of light, including viewpoint-dependent appearance changes caused by light propagation delays to the camera. We render a range of complex effects, including scattering, specular reflection, refraction, and diffraction. Additionally, we demonstrate removing viewpoint-dependent propagation delays using a time warping procedure, rendering of relativistic effects, and video synthesis of direct and global components of light transport.

Video Snapshot Compressive Imaging (SCI) aims to use a low-speed 2D camera to capture high-speed scene as snapshot compressed measurements, followed by a reconstruction algorithm to reconstruct the high-speed video frames. State-of-the-art (SOTA) deep learning-based algorithms have achieved impressive performance, yet with heavy computational workload. Network quantization is a promising way to reduce computational cost. However, a direct low-bit quantization will bring large performance drop. To address this challenge, in this paper, we propose a simple low-bit quantization framework (dubbed Q-SCI) for the end-to-end deep learning-based video SCI reconstruction methods which usually consist of a feature extraction, feature enhancement and video reconstruction module. Specifically, we first design a high-quality feature extraction module and a precise video reconstruction module to extract and propagate high-quality features in the low-bit quantized model. In addition, to alleviate the information distortion of the Transformer branch in the quantized feature enhancement module, we introduce a shift operation on the query and key distributions to further bridge the performance gap. Comprehensive experimental results manifest that our Q-SCI framework achieves superior performance, e.g., Q-SCI (4-bit) can theoretically accelerate previous SOTA real-valued EfficientSCI-S by 7.8× with only 2.3% performance gap on the simulation testing datasets. Code of this paper …

While exploring visual scenes, humans' scanpaths are driven by their underlying attention processes. Understanding visual scanpaths is essential for various applications. Traditional scanpath models predict the where and when of gaze shifts without providing explanations, creating a gap in understanding the rationale behind fixations. To bridge this gap, we introduce GazeXplain, a novel study of visual scanpath prediction and explanation. This involves annotating natural-language explanations for fixations across eye-tracking datasets and proposing a general model with an attention-language decoder that jointly predicts scanpaths and generates explanations. It integrates a unique semantic alignment mechanism to enhance the consistency between fixations and explanations, alongside a cross-dataset co-training approach for generalization. These novelties present a comprehensive and adaptable solution for explainable human visual scanpath prediction. Extensive experiments on diverse eye-tracking datasets demonstrate the effectiveness of GazeXplain in both scanpath prediction and explanation, offering valuable insights into human visual attention and cognitive processes.

Data-driven approaches for autonomous driving (AD) have been widely adopted in the past decade but are confronted with dataset bias and uninterpretability. Inspired by the knowledge-driven nature of human driving, recent approaches explore the potential of large language models (LLMs) to improve understanding and decision-making in traffic scenarios.They find that the pretraining-finetuning paradigm of LLMs on downstream data with the Chain-of-Thought (CoT) reasoning process can enhance explainability and scene understanding. However, such a popular strategy proves to suffer from the notorious problems of misalignment between the crafted CoTs against the consequent decision-making, which remains untouched by previous LLM-based AD methods. To address this problem, we motivate an end-to-end decision-making model based on multimodality-augmented LLM, which simultaneously executes CoT reasoning and carries out planning results. Furthermore, we propose a reasoning-decision alignment constraint between the paired CoTs and planning results, imposing the correspondence between reasoning and decision-making. Moreover, we redesign the CoTs to enable the model to comprehend complex scenarios and enhance decision-making performance. We dub our proposed large language planners with reasoning-decision alignment as RDA-Driver. Experimental evaluations on the nuScenes and DriveLM-nuScnene benchmarks demonstrate the effectiveness of our RDA-Driver in enhancing the performance of end-to-end autonomous driving systems. Specifically, our RDA-Driver …

This paper presents a vector HD-mapping algorithm that formulates the mapping as a tracking task and uses a history of memory latents to ensure consistent reconstructions over time. Our method, MapTracker, accumulates a sensor stream into memory buffers of two latent representations: 1) Raster latents in the bird's-eye-view (BEV) space and 2) Vector latents over the road elements (i.e., road boundaries, dividers, and pedestrian zones). The approach borrows the query propagation paradigm from the tracking literature that explicitly associates tracked road elements from the previous frame to the current, while fusing a subset of memory latents selected with distance strides to further enhance temporal consistency. A vector latent is decoded to reconstruct the geometry of a road element. The paper further makes benchmark contributions by 1) Improving processing code for existing datasets to produce consistent ground truth with temporal alignments and 2) Augmenting existing mAP metrics with consistency checks. MapTracker significantly outperforms existing methods on both nuScenes and Agroverse2 datasets by over 8\% and 19\% on the conventional and the new consistency-aware metrics, respectively. The code and augmented benchmarks will be available.

This paper presents a novel self-supervised two-frame multi-camera metric depth estimation network, termed M^2Depth, which is designed to predict reliable scale-aware surrounding depth in autonomous driving. Unlike the previous works that use multi-view images from a single time-step or multiple time-step images from a single camera, M^2Depth takes temporally adjacent two-frame images from multiple cameras as inputs and produces high-quality surrounding depth. We first construct cost volumes in spatial and temporal domains individually and propose a spatial-temporal fusion module that integrates the spatial-temporal information to yield a strong volume presentation. We additionally combine the neural prior from SAM features with internal features to reduce the ambiguity between foreground and background and strengthen the depth edges. Extensive experimental results on nuScenes and DDAD benchmarks show M^2Depth achieves state-of-the-art performance. The code will be made publicly available.

Vehicle-to-everything (V2X) technology has become an area of interest in research due to the availability of roadside infrastructure perception datasets. However, these datasets primarily focus on urban intersections and lack data on highway scenarios. Additionally, the perception tasks in the datasets are mainly MONO 3D due to limited synchronized data across multiple sensors. To bridge this gap, we propose Highway-V2X (H-V2X), the first large-scale highway Bird's-Eye-View (BEV) perception dataset captured by sensors in the real world. The dataset covers over 100 kilometers of highway, with a diverse range of road and weather conditions. H-V2X consists of over 1.9 million fine-grained categorized samples in BEV space, captured by multiple synchronized cameras, with vector map provided. We performed joint 2D-3D calibrations to ensure correct projection and human labor was involved to ensure data quality. Furthermore, we propose three highly relevant tasks to the highway scenario: BEV detection, BEV tracking, and trajectory prediction. We conducted benchmarks for each task, and innovative methods incorporating vector map information were proposed. We hope that H-V2X and benchmark methods will facilitate highway BEV perception research direction. The dataset and codes will be released upon acceptance.

Quantifying a model’s predictive uncertainty is essential for safety-critical applications such as autonomous driving. We consider quantifying such uncertainty for multi-object detection. In particular, we leverage conformal prediction to obtain uncertainty intervals with guaranteed coverage for object bounding boxes. One challenge in doing so is that bounding box predictions are conditioned on the object's class label. Thus, we develop a novel two-step conformal approach that propagates uncertainty in predicted class labels into the uncertainty intervals for the bounding boxes. This broadens the validity of our conformal coverage guarantees to include incorrectly classified objects, ensuring their usefulness when maximal safety assurances are required. Moreover, we investigate novel ensemble and quantile regression formulations to ensure the bounding box intervals are adaptive to object size, leading to a more balanced coverage across sizes. Validating our two-step approach on real-world datasets for 2D bounding box localization, we find that desired coverage levels are satisfied with actionably tight predictive uncertainty intervals.

We study how vision-language models (VLMs) trained on web-scale data can be integrated into end-to-end driving systems to boost generalization and enable interactivity with human users. While recent approaches adapt VLMs to driving via single-round visual question answering (VQA), human drivers reason about decisions in multiple steps. Starting from the localization of key objects, humans estimate object interactions before taking actions. The key insight is that with our proposed task, Graph VQA, where we model graph-structured reasoning through perception, prediction and planning question-answer pairs, we obtain a suitable proxy task to mimic the human reasoning process. We instantiate datasets (DriveLM-Data) built upon nuScenes and CARLA, and propose a VLM-based baseline approach (DriveLM-Agent) for jointly performing Graph VQA and end-to-end driving. The experiments demonstrate that Graph VQA provides a simple, principled framework for reasoning about a driving scene, and DriveLM-Data provides a challenging benchmark for this task. Our DriveLM-Agent baseline performs end-to-end autonomous driving competitively in comparison to state-of-the-art driving-specific architectures. Notably, its benefits are pronounced when it is evaluated zero-shot on unseen sensor configurations. Our question-wise ablation study shows that the performance gain comes from the rich annotation of prediction and planning QA pairs in the graph structure. To facilitate …

Simulation plays a crucial role in the development of autonomous vehicles (AVs) due to the potential risks associated with real-world testing. Although significant progress has been made in the visual aspects of simulators, generating complex behavior among agents remains a formidable challenge. It is not only imperative to ensure realism in the scenarios generated but also essential to incorporate preferences and conditions to facilitate controllable generation for AV training and evaluation. Traditional methods, which rely mainly on memorizing the distribution of training datasets, often fail to generate unseen scenarios. Inspired by the success of retrieval augmented generation in large language models, we present RealGen, a novel retrieval-based in-context learning framework for traffic scenario generation. RealGen synthesizes new scenarios by combining behaviors from multiple retrieved examples in a gradient-free way, which may originate from templates or tagged scenarios. This in-context learning framework endows versatile generative capabilities, including the ability to edit scenarios, compose various behaviors, and produce critical scenarios. Evaluations show that RealGen offers considerable flexibility and controllability, marking a new direction in the field of controllable traffic scenario generation. Check our project website for more information: https://realgen.github.io/.

Predicting vectorized high-definition (HD) map online is useful for autonomous driving, providing detailed geometric and semantic information on the surrounding road environment. In this paper, we introduce Mask2Map, a novel end-to-end online HD map construction method. Our approach identifies semantic components within a scene represented in the bird's eye view (BEV) domain and then generates a precise vectorized map topology based on this information. Mask2Map comprises two main components: an Instance-level Mask Prediction Network (IMPNet) and a Mask-Driven Map Prediction Network (MMPNet). IMPNet generates a mask-aware query capable of producing BEV segmentation masks, while MMPNet accurately constructs vectorized map components, leveraging the semantic geometric information provided by the mask-aware query. For enhancing HD map predictions, we design innovative modules for MMPNet based on outputs from IMPNet. We present a Positional Feature Generator that generates instance-level positional features by utilizing the comprehensive spatial context from semantic components of instance. We also propose a Geometric Feature Extractor which extracts point-level geometric features using sparse key points pooled from the segmentation masks. Furthermore, we present the denoising training strategy for inter-network consistency to boost the performance of map construction. Our evaluation conducted on nuScenes and Argoverse2 benchmarks demonstrates that our Mask2Map achieves a …

3D dense captioning is a task to localize objects and generate descriptions for each object in a 3D scene. Recent approaches in 3D dense captioning have attempted to incorporate contextual information by modeling relationships with object pairs or aggregating nearest neighbor features of an object. However, the contextual information constructed in these scenarios is limited in two aspects: first, objects have multiple positional relationship that exists across the entire global scene (not only near the object itself), and second, it faces with contradicting objectives--where localization and attribute descriptions are generated better with tight localization, while descriptions involving global positional relations are generated better with contextualized features of the global scene. To overcome this challenge, we introduce CSI, a transformer encoder-decoder pipeline that engages in 3D dense captioning for each object with contextualized geometries (where the structural contexts relevant to each object is summarized) and contextualized objects (where the objects relevant to the summarized structural contexts are aggregated). This simple extension relieves previous methods from the contradicting objectives, enhancing both localization performance while enabling to aggregate contextual features throughout the global scene; thus improving caption generation performance simultaneously. Extensive experiments on two of the most widely-used 3D dense captioning datasets (ScanRefer …

Scene Graph Generation (SGG) offers a structured representation critical in many computer vision applications.Traditional SGG approaches, however, are limited by a closed-set assumption, restricting their ability to recognize only predefined object and relation categories. To overcome this, we categorize SGG scenarios into four distinct settings based on the node and edge: Closed-set SGG, Open Vocabulary (object) Detection-based SGG (OvD-SGG), Open Vocabulary Relation-based SGG (OvR-SGG), and Open Vocabulary Detection + Relation-based SGG (OvD+R-SGG). While object-centric open vocabulary SGG has been studied recently, the more challenging problem of relation-involved open-vocabulary SGG remains relatively unexplored. To fill this gap, we propose a unified framework named OvSGTR towards fully open vocabulary SGG from a holistic view. The proposed framework is an end-to-end transformer architecture, which learns a visual-concept alignment for both nodes and edges, enabling the model to recognize unseen categories. For the more challenging settings of relation-involved open vocabulary SGG, the proposed approach integrates relation-aware pre-training utilizing image-caption data and retains visual-concept alignment through knowledge distillation. Comprehensive experimental results on the Visual Genome benchmark demonstrate the effectiveness and superiority of the proposed framework.

Class-agnostic counting methods enumerate objects of an arbitrary class, providing tremendous utility in many fields. Prior works have limited usefulness as they require either a set of examples of the type to be counted or that the query image contains only a single type of object. A significant factor in these shortcomings is the lack of a dataset to properly address counting in settings with more than one kind of object present. To address these issues, we propose the first Multi-class, Class-Agnostic Counting dataset (MCAC) and A Blind Counter (ABC123), a method that can count multiple types of objects simultaneously without using examples of type during training or inference. ABC123 introduces a new paradigm where instead of requiring exemplars to guide the enumeration, examples are found after the counting stage to help a user understand the generated outputs. We show that ABC123 outperforms contemporary methods on MCAC without needing human in-the-loop annotations. We also show that this performance transfers to FSC-147, the standard class-agnostic counting dataset. MCAC is available at MCAC.active.vision and ABC123 is available at ABC123.active.vision.

Spatio-temporal scene graphs represent interactions in a video by decomposing scenes into individual objects and their pair-wise temporal relationships. Long-term anticipation of the fine-grained pair-wise relationships between objects is a challenging problem. To this end, we introduce the task of Scene Graph Anticipation (SGA). We adopt state-of-the-art scene graph generation methods as baselines to anticipate future pair-wise relationships between objects. In our proposed approaches, we leverage object-centric representations of relationships to reason about the observed video frames and model the evolution of relationships between objects. We take a continuous time perspective and model the latent dynamics of the evolution of object interactions using concepts of NeuralODE and NeuralSDE, respectively. We infer representations of future relationships by solving an Ordinary Differential Equation and a Stochastic Differential Equation, respectively. Extensive experimentation on the Action Genome dataset validates the efficacy of the proposed methods.

Category-level articulated object pose estimation task focuses on the pose estimation of unknown articulated objects within known categories. Despite its significance, this task remains challenging due to the varying shapes and poses of objects, expensive dataset annotation costs, and complex real-world environments. In this paper, we propose a novel self-supervised approach that leverages a single-frame point cloud to accomplish the aforementioned task. Our model consistently generates canonical reconstruction with a canonical pose and joint state for the entire input object, and it estimates object-level pose that reduce overall pose variance and part-level poses that align each part of the input with its corresponding part of the reconstruction. Experimental results demonstrate our approach achieves state-of-the-art performance to other self-supervised methods and comparable performance to other supervised methods. To assess the performance of our model in real-world scenarios, we also introduce a novel real-world articulated object benchmark dataset.

Computer vision methods that explicitly detect object parts and reason on them are a step towards inherently interpretable models. Existing approaches that perform part discovery driven by a fine-grained classification task make very restrictive assumptions on the geometric properties of the discovered parts; they should be small and compact. Although this prior is useful in some cases, in this paper we show that pre-trained transformer-based vision models, such as self-supervised DINOv2 ViT, enable the relaxation of these constraints. In particular, we find that a total variation (TV) prior, which allows for multiple connected components of any size, substantially outperforms previous work. We test our approach on three fine-grained classification benchmarks: CUB, PartImageNet and Oxford Flowers, and compare our results to previously published methods as well as a re-implementation of the state-of-the-art method PDiscoNet with a transformer-based backbone. We consistently obtain substantial improvements across the board, both on part discovery metrics and the downstream classification task, showing that the strong inductive biases in self-supervised ViT models require to rethink the geometric priors that can be used for unsupervised part discovery.

Vehicle-to-everything (V2X) technology has become an area of interest in research due to the availability of roadside infrastructure perception datasets. However, these datasets primarily focus on urban intersections and lack data on highway scenarios. Additionally, the perception tasks in the datasets are mainly MONO 3D due to limited synchronized data across multiple sensors. To bridge this gap, we propose Highway-V2X (H-V2X), the first large-scale highway Bird's-Eye-View (BEV) perception dataset captured by sensors in the real world. The dataset covers over 100 kilometers of highway, with a diverse range of road and weather conditions. H-V2X consists of over 1.9 million fine-grained categorized samples in BEV space, captured by multiple synchronized cameras, with vector map provided. We performed joint 2D-3D calibrations to ensure correct projection and human labor was involved to ensure data quality. Furthermore, we propose three highly relevant tasks to the highway scenario: BEV detection, BEV tracking, and trajectory prediction. We conducted benchmarks for each task, and innovative methods incorporating vector map information were proposed. We hope that H-V2X and benchmark methods will facilitate highway BEV perception research direction. The dataset and codes will be released upon acceptance.

Simulation plays a crucial role in the development of autonomous vehicles (AVs) due to the potential risks associated with real-world testing. Although significant progress has been made in the visual aspects of simulators, generating complex behavior among agents remains a formidable challenge. It is not only imperative to ensure realism in the scenarios generated but also essential to incorporate preferences and conditions to facilitate controllable generation for AV training and evaluation. Traditional methods, which rely mainly on memorizing the distribution of training datasets, often fail to generate unseen scenarios. Inspired by the success of retrieval augmented generation in large language models, we present RealGen, a novel retrieval-based in-context learning framework for traffic scenario generation. RealGen synthesizes new scenarios by combining behaviors from multiple retrieved examples in a gradient-free way, which may originate from templates or tagged scenarios. This in-context learning framework endows versatile generative capabilities, including the ability to edit scenarios, compose various behaviors, and produce critical scenarios. Evaluations show that RealGen offers considerable flexibility and controllability, marking a new direction in the field of controllable traffic scenario generation. Check our project website for more information: https://realgen.github.io/.

We study how vision-language models (VLMs) trained on web-scale data can be integrated into end-to-end driving systems to boost generalization and enable interactivity with human users. While recent approaches adapt VLMs to driving via single-round visual question answering (VQA), human drivers reason about decisions in multiple steps. Starting from the localization of key objects, humans estimate object interactions before taking actions. The key insight is that with our proposed task, Graph VQA, where we model graph-structured reasoning through perception, prediction and planning question-answer pairs, we obtain a suitable proxy task to mimic the human reasoning process. We instantiate datasets (DriveLM-Data) built upon nuScenes and CARLA, and propose a VLM-based baseline approach (DriveLM-Agent) for jointly performing Graph VQA and end-to-end driving. The experiments demonstrate that Graph VQA provides a simple, principled framework for reasoning about a driving scene, and DriveLM-Data provides a challenging benchmark for this task. Our DriveLM-Agent baseline performs end-to-end autonomous driving competitively in comparison to state-of-the-art driving-specific architectures. Notably, its benefits are pronounced when it is evaluated zero-shot on unseen sensor configurations. Our question-wise ablation study shows that the performance gain comes from the rich annotation of prediction and planning QA pairs in the graph structure. To facilitate …

Data-driven approaches for autonomous driving (AD) have been widely adopted in the past decade but are confronted with dataset bias and uninterpretability. Inspired by the knowledge-driven nature of human driving, recent approaches explore the potential of large language models (LLMs) to improve understanding and decision-making in traffic scenarios.They find that the pretraining-finetuning paradigm of LLMs on downstream data with the Chain-of-Thought (CoT) reasoning process can enhance explainability and scene understanding. However, such a popular strategy proves to suffer from the notorious problems of misalignment between the crafted CoTs against the consequent decision-making, which remains untouched by previous LLM-based AD methods. To address this problem, we motivate an end-to-end decision-making model based on multimodality-augmented LLM, which simultaneously executes CoT reasoning and carries out planning results. Furthermore, we propose a reasoning-decision alignment constraint between the paired CoTs and planning results, imposing the correspondence between reasoning and decision-making. Moreover, we redesign the CoTs to enable the model to comprehend complex scenarios and enhance decision-making performance. We dub our proposed large language planners with reasoning-decision alignment as RDA-Driver. Experimental evaluations on the nuScenes and DriveLM-nuScnene benchmarks demonstrate the effectiveness of our RDA-Driver in enhancing the performance of end-to-end autonomous driving systems. Specifically, our RDA-Driver …

This paper presents a novel self-supervised two-frame multi-camera metric depth estimation network, termed M^2Depth, which is designed to predict reliable scale-aware surrounding depth in autonomous driving. Unlike the previous works that use multi-view images from a single time-step or multiple time-step images from a single camera, M^2Depth takes temporally adjacent two-frame images from multiple cameras as inputs and produces high-quality surrounding depth. We first construct cost volumes in spatial and temporal domains individually and propose a spatial-temporal fusion module that integrates the spatial-temporal information to yield a strong volume presentation. We additionally combine the neural prior from SAM features with internal features to reduce the ambiguity between foreground and background and strengthen the depth edges. Extensive experimental results on nuScenes and DDAD benchmarks show M^2Depth achieves state-of-the-art performance. The code will be made publicly available.

This paper presents a vector HD-mapping algorithm that formulates the mapping as a tracking task and uses a history of memory latents to ensure consistent reconstructions over time. Our method, MapTracker, accumulates a sensor stream into memory buffers of two latent representations: 1) Raster latents in the bird's-eye-view (BEV) space and 2) Vector latents over the road elements (i.e., road boundaries, dividers, and pedestrian zones). The approach borrows the query propagation paradigm from the tracking literature that explicitly associates tracked road elements from the previous frame to the current, while fusing a subset of memory latents selected with distance strides to further enhance temporal consistency. A vector latent is decoded to reconstruct the geometry of a road element. The paper further makes benchmark contributions by 1) Improving processing code for existing datasets to produce consistent ground truth with temporal alignments and 2) Augmenting existing mAP metrics with consistency checks. MapTracker significantly outperforms existing methods on both nuScenes and Agroverse2 datasets by over 8\% and 19\% on the conventional and the new consistency-aware metrics, respectively. The code and augmented benchmarks will be available.

Quantifying a model’s predictive uncertainty is essential for safety-critical applications such as autonomous driving. We consider quantifying such uncertainty for multi-object detection. In particular, we leverage conformal prediction to obtain uncertainty intervals with guaranteed coverage for object bounding boxes. One challenge in doing so is that bounding box predictions are conditioned on the object's class label. Thus, we develop a novel two-step conformal approach that propagates uncertainty in predicted class labels into the uncertainty intervals for the bounding boxes. This broadens the validity of our conformal coverage guarantees to include incorrectly classified objects, ensuring their usefulness when maximal safety assurances are required. Moreover, we investigate novel ensemble and quantile regression formulations to ensure the bounding box intervals are adaptive to object size, leading to a more balanced coverage across sizes. Validating our two-step approach on real-world datasets for 2D bounding box localization, we find that desired coverage levels are satisfied with actionably tight predictive uncertainty intervals.

Video Snapshot Compressive Imaging (SCI) aims to use a low-speed 2D camera to capture high-speed scene as snapshot compressed measurements, followed by a reconstruction algorithm to reconstruct the high-speed video frames. State-of-the-art (SOTA) deep learning-based algorithms have achieved impressive performance, yet with heavy computational workload. Network quantization is a promising way to reduce computational cost. However, a direct low-bit quantization will bring large performance drop. To address this challenge, in this paper, we propose a simple low-bit quantization framework (dubbed Q-SCI) for the end-to-end deep learning-based video SCI reconstruction methods which usually consist of a feature extraction, feature enhancement and video reconstruction module. Specifically, we first design a high-quality feature extraction module and a precise video reconstruction module to extract and propagate high-quality features in the low-bit quantized model. In addition, to alleviate the information distortion of the Transformer branch in the quantized feature enhancement module, we introduce a shift operation on the query and key distributions to further bridge the performance gap. Comprehensive experimental results manifest that our Q-SCI framework achieves superior performance, e.g., Q-SCI (4-bit) can theoretically accelerate previous SOTA real-valued EfficientSCI-S by 7.8× with only 2.3% performance gap on the simulation testing datasets. Code of this paper …

Single-photon cameras (SPCs) are emerging as sensors of choice for various challenging imaging applications. One class of SPCs based on the single-photon avalanche diode (SPAD) detects individual photons using an avalanche process; the raw photon data can then be processed to extract scene information under extremely low light, high dynamic range, and rapid motion. Yet, single-photon sensitivity in SPADs comes at a cost --- each photon detection consumes more energy than that of a CMOS camera. This avalanche power significantly limits sensor resolution and could restrict widespread adoption of SPAD-based SPCs. We propose a computational-imaging approach called photon inhibition to address this challenge. Photon inhibition strategically allocates detections in space and time based on downstream inference task goals and resource constraints. We develop lightweight, on-sensor computational inhibition policies that use past photon data to disable SPAD pixels in real-time, to select the most informative future photons. As case studies, we design policies tailored for image reconstruction and edge detection, and demonstrate, both via simulations and real SPC captured data, considerable reduction in photon detections (over 90% of photons) while maintaining task performance metrics. Our work raises the question of ``which photons should be detected?'', and paves the way for future …

Snapshot compressive spectral imaging reconstruction aims to reconstruct three-dimensional spatial-spectral images from a single-shot two-dimensional compressed measurement. Existing state-of-the-art methods are mostly based on deep unfolding structures but have intrinsic performance bottlenecks: i) the ill-posed problem of dealing with heavily degraded measurement, and ii) the regression loss-based reconstruction models being prone to recover images with few details. In this paper, we introduce a generative model, namely the latent diffusion model (LDM), to generate degradation-free prior to enhance the regression-based deep unfolding method by a two-stage training procedure. Furthermore, we propose a Trident Transformer (TT), which extracts correlations among prior knowledge, spatial, and spectral features, to integrate knowledge priors in deep unfolding denoiser, and guide the reconstruction for compensating high-quality spectral signal details. To our knowledge, this is the first approach to integrate physics-driven deep unfolding with generative LDM in the context of CASSI reconstruction. Numeric and visual comparisons on synthetic and real-world datasets illustrate the superiority of our proposed method in both reconstruction quality and computational efficiency. Code will be released.

A minimalist vision system uses the smallest number of pixels needed to solve a vision task. While traditional cameras use a large grid of square pixels, a minimalist camera uses freeform pixels that can take on arbitrary shapes to increase their information content. We show that the hardware of a minimalist camera can be modeled as the first layer of a neural network, where the subsequent layers are used for inference. Training the network for any given task yields the shapes of the camera's freeform pixels, each of which is implemented using a photodetector and an optical mask. We have designed minimalist cameras for monitoring indoor spaces (with 8 pixels), measuring room lighting (with 8 pixels), and estimating traffic flow (with 8 pixels). The performance demonstrated by these systems is on par with a traditional camera with orders of magnitude more pixels. Minimalist vision has two major advantages. First, it naturally tends to preserve the privacy of individuals in the scene since the captured information is inadequate for extracting visual details. Second, since the number of measurements made by a minimalist camera is very small, we show that it can be fully self-powered, i.e., function without an external power supply …

We introduce RAFT2, a faster, simpler, and more accurate RAFT for optical flow. Compared with RAFT, RAFT2 is supervised with a mixture of Laplace loss. It directly regresses an initial flow for faster convergence in recurrent refinements and introduces stereo pretraining to improve generalization. RAFT2 achieves state-of-the-art on Spring benchmark with 3.69 end-point-error (EPE) and 0.36 1-pixel outlier rate (1px), representing 22.9% and 17.8% error reduction from best-published results. In addition, RAFT2 obtains the best cross-dataset generalization on KITTI(train) and Spring(train). With its high efficiency, RAFT2 operates at least 2.3x faster than mainstream methods while maintaining competitive performance, advancing the state of recurrent refinement frameworks in optical flow estimation.

Brain-inspired Spiking Neural Networks (SNNs) have bio-plausibility and low-power advantages over Artificial Neural Networks (ANNs). Applications of SNNs are currently limited to simple classification tasks because of their poor performance. In this work, we focus on bridging the performance gap between ANNs and SNNs on object detection. Our design revolves around network architecture and spiking neuron. First, the overly complex module design causes spike degradation when the YOLO series is converted to the corresponding spiking version. We design a SpikeYOLO architecture to solve this problem by simplifying the vanilla YOLO and incorporating meta SNN blocks. Second, object detection is more sensitive to quantization errors in the conversion of membrane potentials into binary spikes by spiking neurons. To address this challenge, we design a new spiking neuron that activates Integer values during training while maintaining spike-driven by extending virtual timestep during inference. The proposed method is validated on both static and neuromorphic object detection datasets. On the static COCO dataset, we obtain 66.2% mAP@50 and 48.9% mAP@50:95, which is +15.0% and +18.7% higher than the prior state-of-the-art SNN, respectively. On the neuromorphic Gen1 dataset, we achieve 67.2% mAP@50, which is +8.2% and +2.5% greater than the existing best SNN model and …

We propose OmniNOCS: a large-scale monocular dataset with 3D Normalized Object Coordinate Space (NOCS), object masks, and 3D bounding box annotations for indoor and outdoor scenes. OmniNOCS has 20 times more object classes and 200 times more instances than existing NOCs datasets (NOCS-Real275, Wild6D). We use OmniNOCS to train a novel, transformer-based monocular NOCS prediction model (NOCSformer) that can predict accurate NOCS, instance masks and poses from 2D object detections across diverse classes. It is the first NOCS model that can generalize to a broad range of classes when prompted with 2D boxes. We evaluate our model on the task of 3D oriented bounding box prediction, where it achieves comparable results to state-of-the-art 3D detection methods such as Cube R-CNN. Unlike other 3D detection methods, our model also provides detailed and accurate 3D object shape and segmentation. We propose a novel benchmark for the task of NOCS prediction based on OmniNOCS, which we hope will serve as a useful baseline for future work in this area. Our dataset and code are at the project website: https://omninocs.github.io

Audio-driven 3D facial animation aims to map input audio to realistic facial motion. Despite significant progress, limitations arise from inconsistent 3D annotations, restricting previous models to training on specific annotations and thereby constraining the training scale. In this work, we present UniTalker, a unified model featuring a multi-head architecture designed to effectively leverage datasets with varied annotations. To enhance training stability and ensure consistency among multi-head outputs, we employ three training strategies, namely, PCA, model warm-up, and pivot identity embedding. To expand the training scale and diversity, we assemble A2F-Bench, comprising five publicly available datasets and three newly curated datasets. These datasets contain a wide range of audio domains, covering multilingual speech voices and songs, thereby scaling the training data from commonly employed datasets, typically less than 1 hour, to 18.5 hours. With a single trained UniTalker model, we achieve substantial lip vertex error reductions of 9.2% for BIWI dataset and 13.7% for VOCASET. Additionally, the pre-trained UniTalker exhibits promise as the foundation model for audio-driven facial animation tasks. Fine-tuning the pre-trained UniTalker on seen datasets further enhances performance on each dataset, with an average error reduction of 6.3% on A2F-Bench. Moreover, fine-tuning UniTalker on an unseen dataset with only …

4D head capture aims to generate dynamic topological meshes and corresponding texture maps from videos, which is widely utilized in movies and games for its ability to simulate facial muscle movements and recover dynamic textures in pore-squeezing. The industry often adopts the method involving multi-view stereo and non-rigid alignment. However, this approach is prone to errors and heavily reliant on time-consuming manual processing by artists. To simplify this process, we propose Topo4D, a novel framework for automatic geometry and texture generation, which optimizes densely aligned 4D heads and 8K texture maps directly from calibrated multi-view time-series images. Specifically, we first represent the time-series faces as a set of dynamic 3D Gaussians with fixed topology in which the Gaussian centers are bound to the mesh vertices. Afterward, we perform alternative geometry and texture optimization frame-by-frame for high-quality geometry and texture learning while maintaining temporal topology stability. Finally, we can extract dynamic facial meshes in regular wiring arrangement and high-fidelity textures with pore-level details from the learned Gaussians. Extensive experiments show that our method achieves superior results than the current SOTA face reconstruction methods both in the quality of meshes and textures. The code will be released publicly.

We introduce a novel framework for 3D human avatar generation and personalization, leveraging text prompts to enhance user engagement and customization. Central to our approach are key innovations aimed at overcoming the challenges in photo-realistic avatar synthesis. Firstly, we utilize a conditional Neural Radiance Fields (NeRF) model, trained on a large-scale unannotated multi-view dataset, to create a versatile initial solution space that accelerates and diversifies avatar generation. Secondly, we develop a geometric prior, leveraging the capabilities of Text-to-Image Diffusion Models, to ensure superior view invariance and enable direct optimization of avatar geometry. These foundational ideas are complemented by our optimization pipeline built on Variational Score Distillation (VSD), which mitigates texture loss and over-saturation issues. As supported by our extensive experiments, these strategies collectively enable the creation of custom avatars with unparalleled visual quality and better adherence to input text prompts.

In this paper, we develop a personalized video relighting algorithm that produces high-quality and temporally consistent relit videos under any pose, expression, and lighting condition in real-time. Existing relighting algorithms typically rely either on publicly available synthetic data, which yields poor relighting results or on actual light stage data which is difficult to acquire. We show that by just capturing recordings of a user watching YouTube videos on a monitor we can train a personalized algorithm capable of performing high-quality relighting under any condition. Our key contribution is a novel image-based neural relighting architecture that effectively separates the intrinsic appearance features - the geometry and reflectance of the face - from the source lighting and then combines them with the target lighting to generate a relit image. This neural architecture enables smoothing of intrinsic appearance features leading to temporally stable video relighting. Both qualitative and quantitative evaluations show that our architecture improves portrait image relighting quality and temporal consistency over state-of-the-art approaches on both casually captured Light Stage at Your Desk' (LSYD) and light-stage-capturedOne Light At a Time' (OLAT) datasets.

Current deep learning-based low-light image enhancement methods often struggle with high-resolution images, and fail to meet the practical demands of visual perception across diverse and unseen scenarios. In this paper, we introduce a novel approach termed CoLIE, which redefines the enhancement process through mapping the 2D coordinates of an underexposed image to its illumination component, conditioned on local context. We propose a reconstruction of enhanced-light images within the HSV space utilizing an implicit neural function combined with an embedded guided filter, thereby significantly reducing computational overhead. Moreover, we introduce a single image-based training loss function to enhance the model's adaptability to various scenes, further enhancing its practical applicability. Through rigorous evaluations, we analyze the properties of our proposed framework, demonstrating its superiority in both image quality and scene adaptability. Furthermore, our evaluation extends to applications in downstream tasks within low-light scenarios, underscoring the practical utility of CoLIE.

Under-display camera (UDC) image restoration aims to restore images distorted by the OLED display panel covering the frontal camera on a smartphone. Previous deep learning-based UDC restoration methods focused on restoring the image within the RGB domain with the collection of real or synthetic RGB datasets. However, UDC images in these datasets exhibit domain differences from real commercial smartphone UDC images while inherently constraining the problem and solution within the RGB domain. To address this issue, we collect well-aligned sensor-level real UDC images using panels from two commercial smartphones equipped with UDC. We also propose a new UDC restoration method to exploit the disparities between degradations caused by different panels, considering that UDC degradations are specific to the type of OLED panel. For this purpose, we train an encoder with an unsupervised learning scheme using triplet loss that aims to extract the inherent degradations caused by different panels from degraded UDC images as implicit representations. The learned panel-specific degradation representations are then provided as priors to our restoration network based on an efficient Transformer network. Extensive experiments show that our proposed method achieves state-of-the-art performance on our real raw image dataset and generalizes well to previous datasets. Our dataset and …

Holography enables intriguing microscopic imaging modalities, particularly through Quantitative Phase Imaging (QPI), which utilizes the phase of coherent light as a way to reveal the contrast in transparent and thin microscopic specimens. Despite the limitation of image sensors, which detect only light intensity, phase information can still be recorded within a two-dimensional interference pattern between two distinct light waves. Numerical reconstruction is later needed to retrieve the amplitude and phase from such holographic measurements. To this end, we introduce HoloADMM, a novel interpretable, learning-based approach for in-line holographic image reconstruction. HoloADMM enhances imaging capability with spatial image super-resolution, offering a versatile framework that accommodates multiple illumination wavelengths and supports extensive refocusing ranges with up to 10 um precision. Our results indicate a substantial improvement in reconstruction quality over existing methods and demonstrate HoloADMM's effective adaptation to real holographic data captured by our Digital in-line Holographic Microscope (DIHM). This work not only advances holographic imaging techniques but also broadens the potential for non-invasive microscopic analysis applications.

We present an imaging and neural rendering technique that seeks to synthesize videos of light propagating through a scene from novel, moving camera viewpoints. Our approach relies on a new ultrafast imaging setup to capture a first-of-its kind, multi-viewpoint video dataset with picosecond-level temporal resolution. Combined with this dataset, we introduce an efficient neural volume rendering framework based on the transient field. This field is defined as a mapping from a 3D point and 2D direction to a high-dimensional, discrete-time signal that represents time-varying radiance at ultrafast timescales. Rendering with transient fields naturally accounts for effects due to the finite speed of light, including viewpoint-dependent appearance changes caused by light propagation delays to the camera. We render a range of complex effects, including scattering, specular reflection, refraction, and diffraction. Additionally, we demonstrate removing viewpoint-dependent propagation delays using a time warping procedure, rendering of relativistic effects, and video synthesis of direct and global components of light transport.

Recent advances in implicit scene representation enable high-fidelity street view novel view synthesis. However, existing methods optimize a neural radiance field for each scene, relying heavily on dense training images and extensive computation resources. To mitigate this shortcoming, we introduce a new method called Efficient Depth-Guided Urban View Synthesis (EDUS) for fast feed-forward inference and efficient per-scene fine-tuning. Different from prior generalizable methods that infer geometry based on feature matching, EDUS leverages noisy predicted geometric priors as guidance to enable generalizable urban view synthesis from sparse input images. The geometric priors allow us to apply our generalizable model directly in the 3D space, gaining robustness across various sparsity levels. Through comprehensive experiments on the KITTI-360 and Waymo datasets, we demonstrate promising generalization abilities on novel street scenes. Moreover, our results indicate that EDUS achieves state-of-the-art performance in sparse view settings when combined with fast test-time optimization.

Existing methods in neural scene reconstruction utilize the Signed Distance Function (SDF) to model the density function. However, in indoor scenes, the density computed from the SDF for a sampled point may not consistently reflect its real importance in volume rendering, often due to the influence of neighboring objects. To tackle this issue, our work proposes a novel approach for indoor scene reconstruction, which instead parameterizes the density function with the Signed Ray Distance Function (SRDF). Firstly, the SRDF is predicted by the network and transformed to a ray-conditioned density function for volume rendering. We argue that the ray-specific SRDF only considers the surface along the camera ray, from which the derived density function is more consistent to the real occupancy than that from the SDF. Secondly, although SRDF and SDF represent different aspects of scene geometries, their values should share the same sign indicating the underlying spatial occupancy. Therefore, this work introduces a SRDF-SDF consistency loss to constrain the signs of the SRDF and SDF outputs. Thirdly, this work proposes a self-supervised visibility task, introducing the physical visibility geometry to the reconstruction task. The visibility task combines prior from predicted SRDF and SDF as pseudo labels, and contributes to …

Neural Radiance Field (NeRF) is a representation for 3D reconstruction from multi-view images. Despite some recent work showing preliminary success in editing a reconstructed NeRF with diffusion prior, they remain struggling to synthesize reasonable geometry in completely uncovered regions. One major reason is the high diversity of synthetic contents from the diffusion model hinders the radiance field from converging to a crisp and deterministic geometry. Moreover, applying latent diffusion models on real data often yields a textural shift that is incoherent to the image condition due to the auto-encoding error. These two problems are further reinforced with the use of pixel-distance losses. To address these issues, we propose to temper the stochasticity of the diffusion model with per-scene customization and mitigate the textural shift with masked adversarial training. During the analyses, we also found the commonly used pixel and perceptual losses are harmful in the NeRF inpainting task. Through rigorous experiments, our framework yields state-of-the-art NeRF inpainting results on various real-world scenes.

Unsigned distance functions (UDFs) have been a vital representation for open surfaces. With different differentiable renderers, current methods are able to train neural networks to infer a UDF by minimizing the rendering errors on the UDF to the multi-view ground truth. However, these differentiable renderers are mainly handcrafted, which makes them either biased on ray-surface intersections, or sensitive to unsigned distance outliers, or not scalable to large scale scenes. To resolve these issues, we present a novel differentiable renderer to infer UDFs more accurately. Instead of using handcrafted equations, our differentiable renderer is a neural network which is pre-trained in a data-driven manner. It learns how to render unsigned distances into depth images, leading to a prior knowledge, dubbed volume rendering priors. To infer a UDF for an unseen scene from multiple RGB images, we generalize the learned volume rendering priors to map inferred unsigned distances in alpha blending for RGB image rendering. Our results show that the learned volume rendering priors are unbiased, robust, scalable, 3D aware, and more importantly, easy to learn. We evaluate our method on both widely used benchmarks and real scenes, and report superior performance over the state-of-the-art methods.

Novel view synthesis (NVS) aims to generate images at arbitrary viewpoints using multi-view images, and recent insights from neural radiance fields (NeRF) have contributed to remarkable improvements. Recently, studies on generalizable NeRF (G-NeRF) have addressed the challenge of per-scene optimization in NeRFs. The construction of radiance fields on-the-fly in G-NeRF simplifies the NVS process, making it well-suited for real-world applications. Meanwhile, G-NeRF still struggles in representing fine details for a specific scene due to the absence of per-scene optimization, even with texture-rich multi-view source inputs. As a remedy, we propose a Geometry-driven Multi-reference Texture transfer network (GMT) available as a plug-and-play module designed for G-NeRF. Specifically, we propose ray-imposed deformable convolution (RayDCN), which aligns input and reference features reflecting scene geometry. Additionally, the proposed texture preserving transformer (TPFormer) aggregates multi-view source features while preserving texture information. Consequently, our module enables direct interaction between adjacent pixels during the image enhancement process, which is deficient in G-NeRF models with an independent rendering process per pixel. This addresses constraints that hinder the ability to capture high-frequency details. Experiments show that our plug-and-play module consistently improves G-NeRF models on various benchmark datasets. The code will be publicly available soon.

Rendering realistic images from 3D reconstruction is an essential task of many Computer Vision and Robotics pipelines, notably for mixed-reality applications as well as training autonomous agents in simulated environments. However, the quality of novel views heavily depends of the source reconstruction which is often imperfect due to noisy or missing geometry and appearance. Inspired by the recent success of reference-based super-resolution networks, we propose MaRINeR, a refinement method that leverages information of a nearby mapping image to improve the rendering of a target viewpoint. We first establish matches between the raw rendered image of the scene geometry from the target viewpoint and the nearby reference based on deep features, followed by hierarchical detail transfer. We show improved renderings in quantitative metrics and qualitative examples from both explicit and implicit scene representations. We further employ our method on the downstream tasks of pseudo-ground-truth validation, synthetic data enhancement and detail recovery for renderings of reduced 3D reconstructions.

Recent neural implicit representations (NIRs) have achieved great success in the tasks of 3D reconstruction and novel view synthesis. However, they require the images of a scene from different camera views to be available for one-time training. This is expensive especially for scenarios with large-scale scenes and limited data storage. In view of this, we explore the task of incremental learning for NIRs in this work. We design a student-teacher framework to mitigate the catastrophic forgetting problem. As a result, the student network is able to learn new incoming data while preserving old knowledge simultaneously, aided by a random inquirer and an uncertainty-based filter for effective knowledge merging. Our proposed method is general and thus can be adapted to different implicit representations such as neural radiance field (NeRF) and neural SDF. Extensive experimental results for both 3D reconstruction and novel view synthesis demonstrate the effectiveness of our approach compared to different baselines.

Multi-view 3D surface reconstruction using neural implicit representations has made notable progress by modeling the geometry and view-dependent radiance fields within a unified framework. However, their effectiveness in reconstructing objects with specular or complex surfaces is typically biased by the directional parameterization used in their view-dependent radiance network. Viewing direction and reflection direction are the two most commonly used directional parameterizations but have their own limitations. Typically, utilizing the viewing direction usually struggles to correctly decouple the geometry and appearance of objects with highly specular surfaces, while using the reflection direction tends to yield overly smooth reconstructions for concave or complex structures. In this paper, we analyze their failed cases in detail and propose a novel hybrid directional parameterization to address their limitations in a unified form. Extensive experiments demonstrate the proposed hybrid directional parameterization consistently delivered satisfactory results in reconstructing objects with a wide variety of materials, geometry and appearance, whereas using other directional parameterizations faces challenges in reconstructing certain objects. Moreover, the proposed hybrid directional parameterization is nearly parameter-free and can be effortlessly applied in any existing neural surface reconstruction method.

This paper presents a novel approach for the differentiable rendering of convex polyhedra, addressing the limitations of recent methods that rely on implicit field supervision. Our technique introduces a strategy that combines non-differentiable computation of hyperplane intersection through duality transform with differentiable optimization for vertex positioning with three-plane intersection, enabling gradient-based optimization without the need for 3D implicit fields. This allows for efficient shape representation across a range of applications, from shape parsing to compact mesh reconstruction. This work not only overcomes the challenges of previous approaches but also sets a new standard for representing shapes with convex polyhedra.

Wide-angle lens distortion in portrait photography presents a significant challenge for capturing photo-realistic and aesthetically pleasing images. Such distortions are especially noticeable in face regions. To rectify facial distortions for a more natural appearance, we propose encapsulating the generative face prior derived from pre-trained StyleGAN. This prior is then leveraged to facilitate the correction of facial regions. For the non-face background, a notable central symmetry relationship exists in the wide-angle imaging process, yet it has not been explored in the correction process. This geometry prior motivates us to introduce a novel constraint with the explicit aim of preserving symmetry throughout the correction process, thereby contributing to a more visually appealing and natural correction in the non-face region. Experiments demonstrate that our approach outperforms previous methods by a large margin, excelling not only in quantitative measures such as line straightness and shape consistency metrics but also in terms of perceptual visual quality. Our source code and model will be made publicly available.

We present an inverse image-formation module that can enhance the robustness of existing visual SLAM pipelines for in-the-wild scenarios. Casual video captures often suffer from motion blur and varying appearances, which degrades the final quality of coherent 3D visual representation. We propose integrating the physical imaging into the SLAM system, which employs linear HDR radiance maps to collect measurements. Specifically, individual frames aggregate images of multiple poses along the camera trajectory to explain prevalent motion blur in hand-held videos. Additionally, we accommodate per-frame appearance variation by dedicating explicit variables for image formation steps, namely white balance, exposure time, and camera response function. Through joint optimization of additional variables, the SLAM pipeline produces high-quality images with more accurate trajectories. Extensive experiments demonstrate that our approach can be incorporated into recent visual SLAM pipelines using various scene representations, such as neural radiance fields or Gaussian splatting.

Objects undergo varying amounts of perspective distortion as they move across a camera's field of view. Models for predicting 3D from a single image often work with crops around the object of interest and ignore the location of the object in the camera's field of view. We note that ignoring this location information further exaggerates the inherent ambiguity in making 3D inferences from 2D images and can prevent models from even fitting to the training data. To mitigate this ambiguity, we propose Intrinsics-Aware Positional Encoding (KPE), which incorporates information about the location of crops in the image and camera intrinsics. Experiments on three popular 3D-from-a-single-image benchmarks: depth prediction on NYU, 3D object detection on KITTI & nuScenes, and predicting 3D shapes of articulated objects on ARCTIC, show the benefits of KPE.

We present MVSGaussian, a new generalizable 3D Gaussian representation approach derived from Multi-View Stereo (MVS) that can efficiently reconstruct unseen scenes. Specifically, 1) we leverage MVS to encode geometry-aware Gaussian representations and decode them into Gaussian parameters. 2) To further enhance performance, we propose a hybrid Gaussian rendering that integrates an efficient volume rendering design for novel view synthesis. 3) To support fast fine-tuning for specific scenes, we introduce a multi-view geometric consistent aggregation strategy to effectively aggregate the point clouds generated by the generalizable model, serving as the initialization for per-scene optimization. Compared with previous generalizable NeRF-based methods, which typically require minutes of fine-tuning and seconds of rendering per image, MVSGaussian achieves real-time rendering (300+ FPS) with better synthesis quality for each scene on a single RTX 3090 GPU. Compared with the vanilla 3D-GS, MVSGaussian achieves better novel view synthesis with 13.3× less training computational cost (45s). Extensive experiments on DTU, Real Forward-facing, NeRF Synthetic, and Tanks and Temples datasets validate that MVSGaussian achieves state-of-the-art performance with convincing generalizability, real-time rendering speed, and fast per-scene optimization.

The advancement of real-time 3D scene reconstruction and novel view synthesis has been significantly propelled by 3D Gaussian Splatting (3DGS). However, effectively training large-scale 3DGS and rendering it in real-time across various scales remains challenging. This paper introduces CityGaussian (CityGS), which employs a novel divide-and-conquer training approach and Level-of-Detail (LoD) strategy for efficient large-scale 3DGS training and rendering. Specifically, the global scene prior and adaptive training data selection enables efficient training and seamless fusion. Based on fused Gaussian primitives, we generate different detail levels through compression, and realize fast rendering across various scales through the proposed block-wise detail levels selection and aggregation strategy. Extensive experimental results on large-scale scenes demonstrate that our approach attains state-of-the-art rendering quality, enabling consistent real-time rendering of large-scale scenes across vastly different scales.

Implicit Neural Representations (INRs) have proven effective in image representation and compression, offering high visual quality and fast rendering speeds with 10-1000 FPS, assuming sufficient GPU resources are available. However, this requirement often hinders their use on low-end devices with limited memory. In response, we propose a groundbreaking paradigm of image representation and compression by 2D Gaussian Splatting, named GaussianImage. We first introduce 2D Gaussian to represent the image, where each Gaussian has 8 parameters including position, covariance and color. Subsequently, we unveil a novel rendering algorithm based on accumulated summation. Remarkably, our method with a minimum of 3x lower GPU memory usage and 5x faster fitting time not only rivals INRs (e.g., WIRE, I-NGP) in representation performance, but also delivers a faster rendering speed of 1500-2000 FPS regardless of parameter size. Furthermore, we integrate existing vector quantization technique to build an image codec. Experimental results demonstrate that our codec attains rate-distortion performance comparable to compression-based INRs such as COIN and COIN++, while facilitating decoding speeds of approximately 1000 FPS. Additionally, initial proof of concept indicates that our codec surpasses COIN and COIN++ in performance when utilizing partial bits-back coding.

This paper considers the problem of generative novel view synthesis (GNVS), generating novel, plausible views of a scene given a limited number of known views. Here, we propose a set-based generative model that can simultaneously generate multiple, self-consistent new views, conditioned on any number of known views. Our approach is not limited to generating a single image at a time and can condition on zero, one, or more views. As a result, when generating a large number of views, our method is not restricted to a low-order autoregressive generation approach and is better able to maintain generated image quality over large sets of images. We evaluate the proposed model on standard NVS datasets and show that it outperforms the state-of-the-art image-based GNVS baselines. Further, we show that the model is capable of generating sets of camera views that have no natural sequential ordering, like loops and binocular trajectories, and significantly outperforms other methods on such tasks.

Scene-level novel view synthesis (NVS) is fundamental to many vision and graphics applications. Recently, pose-conditioned diffusion models have led to significant progress by extracting 3D information from 2D foundation models, but these methods are limited by the lack of scene-level training data. Common dataset choices either consist of isolated objects (Objaverse), or of object-centric scenes with limited pose distributions (DTU, CO3D). In this paper, we create a large-scale scene-level dataset from Internet photo collections, called MegaScenes, which contains over 100K SfM reconstructions from around the world. Internet photos represent a scalable data source but come with challenges such as lighting and transient objects. We address these issues to further create a subset suitable for the task of NVS. Additionally, we analyze failure cases of state-of-the-art NVS methods and significantly improve generation consistency. Through extensive experiments we validate the effectiveness of both our dataset and method on generating in-the-wild scenes.

Recent advances in diffusion models have enabled 3D generation from a single image. However, current methods often produce suboptimal results for novel views, with blurred textures and deviations from the reference image, limiting their practical applications. In this paper, we introduce HiFi-123, a method designed for high-fidelity and multi-view consistent 3D generation. Our contributions are twofold: First, we propose a Reference-Guided Novel View Enhancement (RGNV) technique that significantly improves the fidelity of diffusion-based zero-shot novel view synthesis methods. Second, capitalizing on the RGNV, we present a novel Reference-Guided State Distillation (RGSD) loss. When incorporated into the optimization-based image-to-3D pipeline, our method significantly improves 3D generation quality, achieving state-of-the-art performance. Comprehensive evaluations demonstrate the effectiveness of our approach over existing methods, both qualitatively and quantitatively. Video comparisons are available on the supplementary project page. We will release our code to the public.

The advent of 3D Gaussian Splatting (3DGS) has revolutionized 3D editing, offering efficient, high-fidelity rendering and enabling precise local manipulations. Currently, diffusion-based 2D editing models are harnessed to modify multi-view rendered images, which then guide the editing of 3DGS models. However, this approach faces a critical issue of multi-view inconsistency, where the guidance images exhibit significant discrepancies across views, leading to mode collapse and visual artifacts of 3DGS. To this end, we introduce View-consistent Editing (VcEdit), a novel framework that seamlessly incorporates 3DGS into image editing processes, ensuring multi-view consistency in edited guidance images and effectively mitigating mode collapse issues. VcEdit employs two innovative consistency modules: the Cross-attention Consistency Module and the Editing Consistency Module, both designed to reduce inconsistencies in edited images. By incorporating these consistency modules into an iterative pattern, VcEdit proficiently resolves the issue of multi-view inconsistency, facilitating high-quality 3DGS editing across a diverse range of scenes.

3D generation has witnessed significant advancements, yet efficiently producing high-quality 3D assets from a single image remains challenging. In this paper, we present a triplane autoencoder, which encodes 3D models into a compact triplane latent space to effectively compress both the 3D geometry and texture information. Within the autoencoder framework, we introduce a 3D-aware cross-attention mechanism, which utilizes low-resolution latent representations to query features from a high-resolution 3D feature volume, thereby enhancing the representation capacity of the latent space. Subsequently, we train a diffusion model on this refined latent space. In contrast to solely relying on image embedding for 3D generation, our proposed method advocates for the simultaneous utilization of both image embedding and shape embedding as conditions. Specifically, the shape embedding is estimated via a diffusion prior model conditioned on the image embedding. Through comprehensive experiments, we demonstrate that our method outperforms state-of-the-art algorithms, achieving superior performance while requiring less training data and time. Our approach enables the generation of high-quality 3D assets in merely 7 seconds on a single A100 GPU. More results and visualization can be found on our project page: https://compress3d.github.io/

Synthesizing realistic 3D indoor scenes is a challenging task that traditionally relies on manual arrangement and annotation by expert designers. Recent advances in autoregressive models have automated this process, but they often lack semantic understanding of the relationships and hierarchies present in real-world scenes, yielding limited performance. In this paper, we propose Forest2Seq, a framework that formulates indoor scene synthesis as an order-aware sequential learning problem. Forest2Seq organizes the inherently unordered collection of scene objects into structured, ordered hierarchical scene trees and forests. By employing a clustering-based algorithm and a breadth-first traversal, Forest2Seq derives meaningful orderings and utilizes a transformer to generate realistic 3D scenes autoregressively. Experimental results on standard benchmarks demonstrate Forest2Seq's superiority in synthesizing more realistic scenes compared to top-performing baselines, with significant improvements in FID and KL scores. Our additional experiments for downstream tasks and ablation studies also confirm the importance of incorporating order as a prior in 3D scene generation.

Pixel-aligned implicit models, such as Multi-view PIFu, DeepMultiCap, DoubleField, and SeSDF, are well-established methods for reconstructing a clothed human from sparse views. However, given V images, these models would only combine features from these images in a point-wise and localized manner. In other words, the V images are processed individually and are only combined in a very narrow fashion at the end of the pipeline. To a large extent, this defeats the purpose of having multi-view information since the multi-view task in question is predominantly treated as a single-view task. To resolve this, we introduce 3DFG-PIFu, a pixel-aligned implicit model that exploits multi-view information right from the start and all the way to the end of the pipeline. Our 3DFG-PIFu makes use of 3D Feature Grids to combine features from V images in a global manner (rather than point-wise or localized) and throughout the pipeline. Other than the 3D Feature Grids, 3DFG-PIFu also proposes an iterative mechanism that refines and updates an existing output human mesh using the different views. Moreover, 3DFG-PIFu introduces SDF-based SMPL-X features, which is a new method of incorporating a SMPL-X mesh into a pixel-aligned implicit model. Our experiments show that 3DFG-PIFu significantly outperforms SOTA models …

Existing UV mapping algorithms are designed to operate on well-behaved meshes, instead of the geometry representations produced by state-of-the-art 3D reconstruction and generation techniques. As such, applying these methods to the volume densities recovered by neural radiance fields and related techniques (or meshes triangulated from such fields) results in texture atlases that are too fragmented to be useful for tasks such as view synthesis or appearance editing. We present a UV mapping method designed to operate on geometry produced by 3D reconstruction and generation techniques. Instead of computing a mapping defined on a mesh's vertices, our method Nuvo uses a neural field to represent a continuous UV mapping, and optimizes it to be a valid and well-behaved mapping for just the set of visible points, i.e. only points that affect the scene's appearance. We show that our model is robust to the challenges posed by ill-behaved geometry, and that it produces editable UV mappings that can represent detailed appearance.

We propose a novel image editing technique that enables 3D manipulations on single images, such as object rotation. Existing approaches typically rely on synthetic multi-view datasets for training specialized models, thus constraining their effectiveness on open-domain images featuring significantly more varied layouts and styles. In contrast, our method directly leverages powerful image diffusion models trained on a broad spectrum of text-image pairs and thus retain their exceptional generalization abilities. This objective is realized through the development of an iterative novel view synthesis and geometry alignment algorithm. The algorithm harnesses diffusion models for dual purposes: they provide appearance prior by predicting novel views of the selected object using estimated depth maps, and they act as a geometry critic by correcting misalignments in 3D shapes across the sampled views. Our method can generate high-quality 3D-aware image edits with large viewpoint transformations and high appearance and shape consistency with the input image, pushing the boundaries of what is possible with single-image 3D-aware editing.

Graphics design is important for various applications, including movie production and game design. To create a high-quality scene, designers usually need to spend hours in software like Blender, in which they might need to interleave and repeat operations, such as connecting material nodes, hundreds of times. Moreover, slightly different design goals may require completely different sequences, making automation difficult. In this paper, we propose a system that leverages Vision-Language Models (VLMs), like GPT-4V, to intelligently search the design action space to arrive at an answer that can satisfy a user's intent. Specifically, we design a vision-based edit generator and state evaluator to work together to find the correct sequence of actions to achieve the goal. Inspired by the role of visual imagination in the human design process, we supplement the visual reasoning capabilities of VLMs with ``imagined'' reference images from image-generation models, providing visual grounding of abstract language descriptions. In this paper, we provide empirical evidence suggesting our system can produce simple but tedious Blender editing sequences for tasks such as editing procedural materials from text and/or reference images, as well as adjusting lighting configurations for product renderings in complex scenes.

Virtual interventions enable the physics-based simulation of device deployment within patient-specific coronary artery anatomy. This framework enables the exploration of alternate scenarios by deploying counterfactual device designs within the same anatomy, revealing critical design factors for patient outcomes. In contrast, our ability to simulate alternate scenarios with anatomic counterfactuals is highly limited. In this study, we investigate how Latent Diffusion Models (LDMs) can custom synthesize coronary anatomy for virtual intervention studies. We introduce several adaptations to enforce anatomic constraints regarding topological validity, local morphological shape, and global skeletal structure. Specifically, we regularize the LDM latent space to reduce topological defects and introduce a conditioning framework based on clinically interpretable and editable coronary morpho-skeletons. We lastly extend diffusion model guidance strategies to the context of morpho-skeletal conditioning and propose a novel guidance method that adaptively updates the guiding condition throughout sampling. Our framework enables the generation and editing of coronary anatomy in a controllable manner, allowing device designers to better explore the relationship

Learning radiance fields (NeRF) with powerful 2D diffusion models has garnered popularity for text-to-3D generation. Nevertheless, the implicit 3D representations of NeRF lack explicit modeling of meshes and textures over surfaces, and such surface-undefined way may suffer from the issues, e.g., noisy surfaces with ambiguous texture details or cross-view inconsistency. To alleviate this, we present DreamMesh, a novel text-to-3D architecture that pivots on well-defined surfaces (triangle meshes) to generate high-fidelity explicit 3D model. Technically, DreamMesh capitalizes on a distinctive coarse-to-fine scheme. In the coarse stage, the mesh is first deformed by text-guided Jacobians and then DreamMesh textures the mesh with an interlaced use of 2D diffusion models in a tuning free manner from multiple viewpoints. In the fine stage, DreamMesh jointly manipulates the mesh and refines the texture map, leading to high-quality triangle meshes with high-fidelity textured materials. Extensive experiments demonstrate that DreamMesh significantly outperforms state-of-the-art text-to-3D methods in faithfully generating 3D content with richer textual details and enhanced geometry.

Due to the lack of large-scale text-3D correspondence data, recent text-to-3D generation works mainly rely on utilizing 2D diffusion models for synthesizing 3D data. Since diffusion-based methods typically require significant optimization time for both training and inference, the use of GAN-based models would still be desirable for fast 3D generation. In this work, we propose Triplane Attention for text-guided 3D generation (TPA3D), an end-to-end trainable GAN-based deep learning model for fast text-to-3D generation. With only 3D shape data and their rendered 2D images observed during training, our TPA3D is designed to retrieve detailed visual descriptions for synthesizing the corresponding 3D mesh data. This is achieved by the proposed attention mechanisms on the extracted sentence and word-level text features. In our experiments, we show that TPA3D generates high-quality 3D textured shapes aligned with fine-grained descriptions, while impressive computation efficiency can be observed.

We present a 3D modeling method which enables end-users to refine or detailize 3D shapes using machine learning, expanding the capabilities of AI-assisted 3D content creation. Given a coarse voxel shape (e.g., one produced with a simple box extrusion tool or via generative modeling), a user can directly “paint” desired target styles representing compelling geometric details, from input exemplar shapes, over different regions of the coarse shape. These regions are then up-sampled into high-resolution geometries which adhere with the painted styles. To achieve such controllable and localized 3D detailization, we build on top of a Pyramid GAN by making it masking-aware. We devise novel structural losses and priors to ensure that our method preserves both desired coarse structures and fine-grained features even if the painted styles are borrowed from diverse sources, e.g., different semantic parts and even different shape categories. Through extensive experiments, we show that our ability to localize details enables novel interactive creative workflows and applications. Our experiments further demonstrate that in comparison to prior techniques built on global detailization, our method generates structure-preserving, high-resolution stylized geometries with more coherent shape details and style transitions.

In this paper, we tackle a new and challenging problem of text-driven generation of 3D garments with high-quality textures. We propose WordRobe, a novel framework for the generation of unposed & textured 3D garment meshes from user-friendly text prompts. We achieve this by first learning a latent representation of 3D garments using a novel coarse-to-fine training strategy and a loss for latent disentanglement, promoting better latent interpolation. Subsequently, we align the garment latent space to the CLIP embedding space in a weakly supervised manner, enabling text-driven 3D garment generation and editing. For appearance modeling, we leverage the zero-shot generation capability of ControlNet to synthesize view-consistent texture maps in a single feed-forward inference step, thereby drastically decreasing the generation time as compared to existing methods. We demonstrate superior performance over current SOTAs for learning 3D garment latent space, garment interpolation, and text-driven texture synthesis, supported by quantitative evaluation and qualitative user study. The unposed 3D garment meshes generated using WordRobe can be directly fed to standard cloth simulation & animation pipelines without any post-processing.

Inspired by cognitive theories, we introduce AnyHome, a framework that translates any text into well-structured and textured indoor scenes at a house-scale. By prompting Large Language Models (LLMs) with designed templates, our approach converts provided textual narratives into amodal structured representations. These representations guarantee consistent and realistic spatial layouts by directing the synthesis of a geometry mesh within defined constraints. A Score Distillation Sampling process is then employed to refine the geometry, followed by an egocentric inpainting process that adds lifelike textures to it. AnyHome stands out with its editability, customizability, diversity, and realism. The structured representations for scenes allow for extensive editing at varying levels of granularity. Capable of interpreting texts ranging from simple labels to detailed narratives, AnyHome generates detailed geometries and textures that outperform existing methods in both quantitative and qualitative measures.

Text-to-image diffusion models have significantly advanced in conditional image generation. However, these models usually struggle with accurately rendering images featuring humans, resulting in distorted limbs and other anomalies. This issue primarily stems from the insufficient recognition and evaluation of limb qualities in diffusion models. To address this issue, we introduce AbHuman, the first large-scale synthesized human benchmark focusing on anatomical anomalies. This benchmark consists of 56K synthesized human images, each annotated with detailed, bounding-box level labels identifying 147K human anomalies in 18 different categories. Based on this, the recognition of human anomalies can be established, which in turn enhances image generation through traditional techniques such as negative prompting and guidance. To further boost the improvement, we propose HumanRefiner, a novel plug-and-play approach for the coarse-to-fine refinement of human anomalies in text-to-image generation. Specifically, HumanRefiner utilizes a self-diagnostic procedure to detect and correct issues related to both coarse-grained abnormal human poses and fine-grained anomaly levels, facilitating pose-reversible diffusion generation. Experimental results on the AbHuman benchmark demonstrate that HumanRefiner significantly reduces generative discrepancies, achieving a 2.9x improvement in limb quality compared to the state-of-the-art open-source generator SDXL and a 1.4x improvement over DALL-E 3 in human evaluations. The dataset and code will …

We present SENC, a novel self-supervised neural cloth simulator that addresses the challenge of cloth self-collision. This problem has remained unresolved due to the gap in simulation setup between recent collision detection and response approaches and self-supervised neural simulators. The former requires collision-free initial setups, while the latter necessitates random cloth instantiation during training. To tackle this issue, we propose a novel loss based on Global Intersection Analysis (GIA). This loss extracts the volume surrounded by the cloth region that forms the penetration. By constructing an energy based on this volume, our self-supervised neural simulator can effectively address cloth self-collisions. Moreover, we develop a self-collision-aware graph neural network capable of learning to handle self-collisions, even for parts that are topologically distant from one another. Additionally, we introduce an effective external force scheme that enables the simulation to learn the cloth's behavior in response to random external forces. We validate the efficacy of SENC through extensive quantitative and qualitative experiments, demonstrating that it effectively reduces cloth self-collision while maintaining high-quality animation results.

Advances in 3D generation have facilitated sequential 3D model generation (a.k.a 4D generation), yet its application for animatable objects with large motion remains scarce. Our work proposes {AnimatableDreamer}, a text-to-4D generation framework capable of generating diverse categories of non-rigid objects on skeletons extracted from a monocular video. At its core, {AnimatableDreamer} is equipped with our novel optimization design dubbed Canonical Score Distillation (CSD), which lifts 2D diffusion for temporal consistent 4D generation. CSD, designed from a score gradient perspective, generates a canonical model with warp-robustness across different articulations. Notably, it also enhances the authenticity of bones and skinning by integrating inductive priors from a diffusion model. Furthermore, with multi-view distillation, CSD infers invisible regions, thereby improving the fidelity of monocular non-rigid reconstruction. Extensive experiments demonstrate the capability of our method in generating high-flexibility text-guided 3D models from the monocular video, while also showing improved reconstruction performance over existing non-rigid reconstruction methods. Project page \url{https://AnimatableDreamer.github.io/}.

We introduce SceneScript,a method that directly produces full scene models as a sequence of structured language commands using an autoregressive, token-based approach. Our proposed scene representation is inspired by recent successes in transformers & LLMs, and departs from more traditional methods which commonly describe scenes as meshes, voxel grids, point clouds or radiance fields. Our method infers the set of structured language commands directly from encoded visual data using a scene language encoder-decoder architecture. To train SceneScript, we generate and release a large-scale synthetic dataset called Anonymized Dataset consisting of 100k high-quality indoor scenes, with photorealistic and ground-truth annotated renders of egocentric scene walkthroughs. Our method gives state-of-the art results in architectural layout estimation, and competitive results in 3D object detection. Lastly, we explore an advantage for SceneScript, which is the ability to readily adapt to new commands via simple additions to the structured language, which we illustrate for tasks such as coarse 3D object part reconstruction.

This paper demonstrates how to use generative models as data mining tools. Our insight is that generative methods learn an accurate model of their training data that we can analyze to summarize and understand that data. This analysis-by-synthesis approach to data mining has two key advantages. First, it scales much better than traditional correspondence-based approaches since it does not require explicitly comparing all pairs of visual elements. Second, generative models can disentangle factors of variation within the training data (such as appearance vs. geographic location), which are nearly always entangled within the data itself. In this work, we train a location-conditioned diffusion model on worldwide streetview data to mine for geographical visual patterns. We synthesize a dataset of parallel images depicting the same scene layouts across different locations using this model. We define typicality measures, assessing how characteristic visual elements are for geographic location, either country-specific or in terms of cross-country variability.

We introduce \emph{ReMatching}, a novel shape correspondence solution based on the functional maps framework. Our method, by exploiting a new and appropriate \emph{re}-meshing paradigm, can target shape-\emph{matching} tasks even on meshes counting millions of vertices, where the original functional maps does not apply or requires a massive computational cost. The core of our procedure is a time-efficient remeshing algorithm which constructs a low-resolution geometry while acting conservatively on the original topology and metric. These properties allow translating the functional maps optimization problem on the resulting low-resolution representation, thus enabling efficient computation of correspondences with functional map approaches. Finally, we propose an efficient technique for extending the estimated correspondence to the original meshes. Through quantitative and qualitative comparisons, we show that our method is more efficient and effective, outperforming state-of-the-art pipelines in quality and computational cost.

Reconstructing the geometry and topology structure from raw unstructured data has always been an important research topic in indoor mapping. In this paper, we aim to reconstruct the floorplan with a vectorized representation from point clouds. Although considerable advancements have been achieved in this field over recent years, current methods still face several challenges, including missing corners or edges, inaccurate corner positions or angles, self-intersecting or overlapping polygons and even implausible topology. To address these challenges, we present PolyRoom, a room-aware Transformer for floorplan reconstruction by introducing uniform sampling representation, room query initialization and hierarchical self-attention. Specifically, we first project the 3D point clouds as a 2D density map and initialize room queries through instance segmentation and uniform sampling. Then, in the Transformer, room queries first interact with each other with hierarchical self-attention. After that, with deformable cross-attention, vertex coordinates are refined layer by layer under dense supervision. Finally, the compact and structured floorplan is extracted through simple corner vertex selection operations. Experimental results on two widely used datasets demonstrate that PolyRoom outperforms current state-of-the-art methods both quantitatively and qualitatively. We will publicly share all our code.

Polygonal mesh reconstruction of a raw point cloud is a valuable topic in the field of computer graphics and 3D vision. Especially to 3D architectural models, polygonal mesh provides concise expressions for fundamental geometric structures while effectively reducing data volume. However, there have some limitations of traditional reconstruction methods: normal vector dependency, noisy points and defective parts sensitivity, and internal geometric structure lost, which reduce the practicality in real scene. In this paper, we propose a robust and efficient polygonal mesh reconstruction method to address the issues in architectural point cloud reconstruction task. It is an iterative adaptation process to detect planar shapes from scattered points. The initial structural polygonal mesh can be established in the constructed convex polyhedral space without assistance of normal vectors. Then, we develop an efficient polygon-based winding number strategy to orient polygonal mesh with global consistency. The significant advantage of our method is to provide a structural reconstruction for architectural point clouds and avoid point-based normal vector analysis. It effectively improves the robustness to noisy points and defective parts. More geometric details can be preserved in the reconstructed polygonal mesh. Experimental results show that our method can reconstruct concise, oriented and faithfully polygonal mesh that …

With the maturity of depth sensors, point clouds have received increasing attention in various 3D safety-critical applications, while deep point cloud learning models have been shown to be vulnerable to adversarial attacks. Most existing 3D attackers rely on implicit global distance losses to perturb whole points, failing to restrict the proper 3D geometry as point clouds are highly structured. To this end, in this paper, we propose a novel Wavelet Patches Attack (WPA), which leverages local spectral attributes to identify curvature-aware patches for hiding imperceptible perturbations aligned with their local geometric characteristics. Specifically, WPA first transforms the point cloud into the spectral domain using a wavelet operator, obtaining potential geometric structures in different local regions. Each wavelet corresponds to different curvature contexts of local points. Then, by decomposing the 3D object with different curvature-aware levels through the wavelet coefficients, we can perceive the local geometric characteristics and get various curvature-consistent patches. At last, based on the curvature variations of patches, WPA introduces two-type perturbations along the tangent plane and normal vector direction to hide imperceptible noise in slow- and fast-variation patches for preserving the geometric-sensitive local characteristics of smoothness and sharpness, respectively. Experiments demonstrate the superior imperceptibility of our attack …

In this paper, we explore a novel framework, EGIInet (Explicitly Guided Information Interaction Network), a model for View-guided Point cloud Completion (ViPC) task, which aims to restore a complete point cloud from a partial one with a single view image. In comparison with previous methods that relied on the global semantics of input images, EGIInet efficiently combines the information from two modalities by leveraging the geometric nature of the completion task. Specifically, we propose an explicitly guided information interaction strategy supported by modal alignment for point cloud completion. First, in contrast to previous methods which simply use 2D and 3D backbones to encode features respectively, we unified the encoding process to promote modal alignment. Second, we propose a novel explicitly guided information interaction strategy that could help the network identify critical information within images, thus achieving better guidance for completion. Extensive experiments demonstrate the effectiveness of our framework, and we achieved a new state-of-the-art (+16% CD over XMFnet) in benchmark datasets despite using fewer parameters than the previous methods. The pre-trained model and code will be released upon acceptance.

In this work, we address the task of point cloud denoising using a novel framework adapting Diffusion Schrödinger bridges to unstructured data like point sets. Unlike previous works that predict point-wise displacements from point features or learned noise distributions, our method learns an optimal transport plan between paired point clouds. In experiments on object datasets such as the PU-Net dataset and real-world datasets like ScanNet++ and ARKitScenes, P2P-Bridge improves by a notable margin over existing methods. Although our method demonstrates promising results utilizing solely point coordinates, we demonstrate that incorporating additional features like RGB information and point-wise DINOV2 features further improves the results. Code will be made public upon acceptance.

Accurate measurement of the discrepancy between point sets is crucial for point cloud learning tasks. Chamfer distance (CD) is favoured over more effective loss metrics such as Earth Mover's Distance (EMD) for this purpose due to its computational efficiency. Previous investigations into loss function improvements almost exclusively focus on 3D losses as static metrics, and ignore their dynamic behaviour during training. We show that directly modifying the correspondence criteria can prevent clustering of points during training, and lead to more uniform point distributions. We propose UniformCD, a novel 3D distance metric that prioritises matching the relative local densities of point neighbourhoods when assigning correspondences. Experiments demonstrate that the proposed loss function improves performance on a variety of tasks such as cloud completion, parametric model optimisation, as well as downstream task performance when used for self-supervised learning, achieving SOTA EMD results among point set objective functions. We show that the proposed method exploits local density information to converge towards globally optimum density distributions, narrowing the disparity between CD and EMD.

Acquiring complete point clouds for real-world scenarios is labor-intensive, making it impractical for conventional learning-based approaches. Numerous methods have been proposed to overcome this limitation by leveraging synthetic complete point clouds. While access to complete point clouds offers a notable advantage, they often struggle to bridge domain gaps, leading to sub-optimal performance. As a remedy, we propose a novel part-based framework for synthetic-to-real domain adaptation in point cloud completion. Our approach starts on the observation that domain gaps inherent in part information are relatively small, as parts are shared properties across categories regardless of domains. To employ part-based approach to point cloud completion, we introduce Part-Based Decomposition (PBD) module to generate part input point clouds. Subsequently, we design a Part-Aware Completion (PAC) module, which operates in a part-wise manner to produce complete point clouds. Within PAC, we devise a novel part-aware transformer to learn relationships between parts and utilize this information to infer missing parts in incomplete point clouds. Extensive experiments demonstrate that our part-based framework significantly outperforms existing studies on real-world point cloud datasets.

The scarcity of annotated data in LiDAR point cloud understanding hinders effective representation learning. Consequently, scholars have been actively investigating efficacious self-supervised pre-training paradigms. Nevertheless, temporal information, which is inherent in the LiDAR point cloud sequence, is consistently disregarded. To better utilize this property, we propose an effective pre-training strategy, namely Temporal Masked Auto-Encoders (T-MAE), which takes as input temporally adjacent frames and learns temporal dependency. A SiamWCA backbone, containing a Siamese encoder and a windowed cross-attention (WCA) module, is established for the two-frame input. Considering that the movement of an ego-vehicle alters the view of the same instance, temporal modeling also serves as a robust and natural data augmentation, enhancing the comprehension of target objects. SiamWCA is a powerful architecture but heavily relies on annotated data. Our T-MAE pre-training strategy alleviates its demand on annotated data. Comprehensive experiments demonstrate that T-MAE achieves the best performance on both Waymo and ONCE datasets among competitive self-supervised approaches.

The complex traffic environment and various weather conditions make the collection of LiDAR data expensive and challenging. Achieving high-quality and controllable LiDAR data generation is urgently needed, controlling with text is a common practice, but there is little research in this field. To this end, we propose Text2LiDAR, the first efficient, diverse, and text-controllable LiDAR data generation model. Specifically, we design an equirectangular transformer architecture, utilizing the designed equirectangular attention to capture LiDAR features in a manner with data characteristics. Then, we design a control-signal embedding injector to efficiently integrate control signals through the global-to-focused attention mechanism. Additionally, we devise a frequency modulator to assist the model in recovering high-frequency details, ensuring the clarity of the generated point cloud. To foster development in the field and optimize text-controlled generation performance, we construct nuLiDARtext which offers diverse text descriptors for 34,149 LiDAR point clouds from 850 scenes. Experiments on uncontrolled and text-controlled generation in various forms on KITTI-360 and nuScenes datasets demonstrate the superiority of our approach.

Progress in 3D computer vision tasks demands a huge amount of data, yet annotating multi-view images with 3D-consistent annotations, or point clouds with part segmentation is both time-consuming and challenging. This paper introduces DatasetNeRF, a novel approach capable of generating infinite, high-quality 3D-consistent 2D annotations alongside 3D point cloud segmentations, while utilizing minimal 2D human-labeled annotations. Specifically, we leverage the strong semantic prior within a 3D generative model to train a semantic decoder, requiring only a handful of fine-grained labeled samples. Once trained, the decoder efficiently generalizes across the latent space, enabling the generation of infinite data. The generated data is applicable across various computer vision tasks, including video segmentation and 3D point cloud segmentation in both synthetic and real-world scenarios. Our approach not only surpasses baseline models in segmentation quality, achieving superior 3D consistency and segmentation precision on individual images, but also demonstrates versatility by being applicable to both articulated and non-articulated generative models. Furthermore, we explore applications stemming from our approach, such as 3D-aware semantic editing and 3D inversion.

The linear inverse problem associated with the standard model for hyperspectral image recovery from CASSI measurements is considered. This is formulated as the minimization of an objective function which is the sum of a total variation regularizer and a least squares loss function. Standard first-order iterative minimization algorithms, such as ISTA, FISTA and TwIST, require as input the value of the Lipschitz constant for the gradient of the loss function, or at least a good upper bound on this value, in order to select appropriate step lengths. For the loss term considered here, this Lipschitz constant equals the square of the largest singular value of the measurement map. In applications, this singular value is usually computed directly as the largest eigenvalue of a huge square matrix. This computation can sometimes be a bottleneck in an otherwise optimized algorithm. In this paper we effectively eliminate this bottleneck for CASSI reconstructions by showing how the Lipschitz constant can be calculated from a square matrix whose size is easily three orders of magnitudes smaller than in the direct approach.

This paper presents a novel approach for reconstructing dynamic radiance fields from monocular videos. We integrate kinematics with dynamic radiance fields, bridging the gap between the sparse nature of monocular videos and the real-world physics. Our method introduces the kinematic field, capturing motion through kinematic quantities: velocity, acceleration, and jerk. The kinematic field is jointly learned with the dynamic radiance field by minimizing the photometric loss without motion ground truth. We further augment our method with physics-driven regularizers grounded in kinematics. We propose physics-driven regularizers that ensure the physical validity of predicted kinematic quantities, including advective acceleration and jerk. Additionally, we control the motion trajectory based on rigidity equations formed with the predicted kinematic quantities. In experiments, our method outperforms the state-of-the-arts by capturing physical motion patterns within challenging real-world monocular videos. Code will be available on GitHub.

Plane adjustment (PA) is important for many 3D applications, which involves simultaneous pose estimation and plane recovery. While significant progress has been made recently, it is still a challenging problem in the realm of multi-view point cloud registration. The successful convergence of current state-of-the-art methods heavily depends on good initialization. Furthermore, the time complexity renders existing approaches impractical for large-scale problems. To address these challenges, we exploit a novel optimization strategy termed Bi-Convex Relaxation for large scale plane adjustment to improve its efficiency, convergence region and scalability. In particular, we decouple the original complex problem into two simpler sub-problems, which are then reformulated using convex relaxation. Subsequently, we can alternately solve these two sub-problems until convergence. On top of this novel optimization strategy, we propose two variants of the algorithm, namely GlobalPointer and GlobalPointer++, based on point-to-plane and plane-to-plane error, respectively. Extensive experiments on both synthetic and real datasets demonstrate that our method is able to perform large-scale plane adjustment with linear time complexity, larger convergence basin, and poor initialization, while achieving similar accuracy as prior methods. We will release our code to the public for further study.

Learning feature correspondence is a foundational task in computer vision, holding immense importance for downstream applications such as visual odometry and 3D reconstruction. Despite recent progress in data-driven models, feature correspondence learning is still limited by the lack of accurate per-pixel correspondence labels. To overcome this difficulty, we introduce a new self-supervised scheme, imperative learning (IL), for training feature correspondence. It enables correspondence learning on arbitrary uninterrupted videos without any camera pose or depth labels, heralding a new era for self-supervised correspondence learning. Specifically, we formulated the problem of correspondence learning as a bilevel optimization, which takes the reprojection error from bundle adjustment as a supervisory signal for the model. To avoid large memory and computation overhead, we leverage the stationary point to effectively back-propagate the implicit gradients through bundle adjustment. Through extensive experiments, we demonstrate superior performance on tasks including feature matching and pose estimation, in which we obtained an average of 30% accuracy gain over the state-of-the-art matching models. The source code will be made public to benefit the community.

We propose a novel approach to the problem of estimating the fundamental matrix from point correspondences and their relative depths. Relative depths can be approximated from the scales of local features, which are commonly available or can be obtained from non-metric monocular depth estimates provided by popular deep learning-based methods. This makes the considered problem very relevant. To derive efficient solutions, we explore new geometric constraints on the fundamental matrix with known relative depths and present new algebraic constraints between the fundamental matrix and the translation vector. Using point correspondences and their relative depths, we derive novel efficient minimal solvers for two fully uncalibrated cameras, two cameras with different unknown focal lengths, and two cameras with equal unknown focal lengths, respectively. We propose different variants of these solvers based on the source of the relative depth information. We present detailed analyses and comparisons with state-of-the-art solvers, including results with 86,306 image pairs from three large-scale datasets.

We propose a novel test-time optimization approach for efficiently and robustly tracking any pixel at any time in a video. The latest state-of-the-art optimization-based tracking technique, OmniMotion, requires a prohibitively long optimization time, rendering it impractical for practical applications. OmniMotion is sensitive to the choice of random seeds, leading to unstable convergence. To improve efficiency and robustness, we introduce a novel invertible deformation network, CaDeX++, which factorizes the function representation into a local spatial-temporal feature grid and enhances the expressivity of the coupling blocks with non-linear functions. While CaDeX++ incorporates a stronger geometric bias within its architectural design, it also takes advantage of the inductive bias provided by the vision foundation models. Our system utilizes monocular depth estimation to represent scene geometry and enhances the objective by incorporating DINOv2 long-term semantics to regulate the optimization process. Our experiments demonstrate a substantial improvement in training speed (more than 10 times faster), robustness, and accuracy in tracking over the SoTA optimization-based method OmniMotion.

This work addresses the challenge of sub-pixel accuracy in detecting 2D local features, a cornerstone problem in computer vision. Despite the advancements brought by neural network-based methods like SuperPoint and ALIKED, these modern approaches lag behind classical ones such as SIFT in keypoint localization accuracy due to their lack of subpixel precision. We propose a novel network that enhances any detector with sub-pixel precision by learning an offset vector for detected features, thereby eliminating the need for designing specialized sub-pixel accurate detectors. This optimization directly minimizes test-time evaluation metrics like relative pose error. Through extensive testing with both nearest neighbors matching and the recent LightGlue matcher across various real-world datasets, our method consistently outperforms existing methods in accuracy. Moreover, it adds only around 7 ms to the time of a particular detector. We will make the code and models public.

Accurate estimation of both the external orientation and internal bending angle is crucial for understanding a soft robot state within its environment. However, existing sensor-based methods face limitations in cost, environmental constraints, and integration issues. Conventional image-based methods struggle with the shape complexity of soft robots. In this paper, we propose a novel shape-guided configuration-aware learning framework for image-based soft robot pose estimation. Inspired by the recent advances in 2D-3D joint representation learning, we leverage the 3D shape prior of the soft robot to enhance its image-based shape representation. Concretely, we first extract the part-level geometry representation of the 3D shape prior, then adapt this representation to the image by querying the image features corresponding to different robot parts. Furthermore, we present an effective mechanism to dynamically deform the shape prior. It aims to mitigate the shape difference between the adopted shape prior and the soft robot depicted in the image. This more expressive shape guidance further boosts the image-based robot representation and can be effectively used for soft robot pose refinement. Extensive experiments on surgical soft robots demonstrate the advantages of our method when compared with a series of keypoint-based, skeleton-based and direct regression-based methods.

Estimating the 6D pose of objects unseen during training is highly desirable yet challenging. Zero-shot object 6D pose estimation methods address this challenge by leveraging additional task-specific supervision provided by large-scale, photo-realistic synthetic datasets. However, their performance heavily depends on the quality and diversity of rendered data and they require extensive training. In this work, we show how to tackle the same task but without training on specific data. We propose FreeZe, a novel solution that harnesses the capabilities of pre-trained geometric and vision foundation models. FreeZe leverages 3D geometric descriptors learned from unrelated 3D point clouds and 2D visual features learned from web-scale 2D images to generate discriminative 3D point-level descriptors. We then estimate the 6D pose of unseen objects by 3D registration based on RANSAC. We also introduce a novel algorithm to solve ambiguous cases due to geometrically symmetric objects that is based on visual features. We comprehensively evaluate FreeZe across the seven core datasets of the BOP Benchmark, which include over a hundred 3D objects and 20,000 images captured in various scenarios. FreeZe consistently outperforms all state-of-the-art approaches, including competitors extensively trained on synthetic 6D pose estimation data. We will release the source code publicly.

6D Object Pose Estimation is a critical yet challenging task in the field of computer vision, distinguished from more traditional 2D tasks by its lack of large-scale datasets. This scarcity hampers comprehensive evaluation of model performance and consequently, limits research development while also restricting the applicability of research across diverse domains due to the limited number of instances or categories available. To address these issues and facilitate progress in this field, this paper introduces Omni6DPose, a substantial dataset characterized by its diversity in object categories, large scale, and variety in object materials. Omni6DPose is divided into three main components: ROPE (Real 6D Object Pose Estimation Dataset), which includes 270,000 images annotated with over one million annotations across 600 instances in 140 categories; SOPE(Simulated 6D Object Pose Estimation Dataset), consisting of 350,000 images created in a mixed reality setting with depth simulation, annotated with over one million annotations across 4,000 instances in the same 140 categories; and the manually aligned real scanned objects used in both ROPE and SOPE. Omni6DPose is inherently challenging due to the substantial variations and ambiguities. To address this challenge, we propose GenPose++, a novel framework incorporating a pre-trained 2D foundational model to enhance generalization capabilities and …

We address the simulation-to-real domain gap in six degree- of-freedom pose estimation (6DoF PE), and propose a novel self-supervised keypoint voting-based 6DoF PE framework, effectively narrowing this gap using a learnable kernel in RKHS. We formulate this domain gap as a distance in high-dimensional feature space, distinct from previ- ous iterative matching methods. We propose an adapter network, that evolves the network parameters from the source domain, which has been trained on synthetic data with synthetic poses, to the target domain, which is trained on real data. Importantly, the real data training only uses pseudo-poses estimated by pseudo-keypoints, and thereby requires no real groundtruth data annotations. Our proposed method is called RKHSPose, and achieves state-of-the-art performance among self-supervised methods on three commonly used 6DoF PE datasets including LINEMOD (+4.2%), Occlusion LINEMOD (+2%), and YCB-Video (+3%). It also compares favorably to fully supervised methods on all six applicable BOP core datasets, achieving within −11.3% to +0.2% of the top fully supervised results.

We introduce a novel bottom-up approach for human body mesh reconstruction, specifically designed to address the challenges posed by partial visibility and occlusion in input images. Traditional top-down methods, relying on whole-body parametric models like SMPL, falter when only a small part of the human is visible, as they require visibility of most of the human body for accurate mesh reconstruction. To overcome this limitation, our method employs a "Divide and Fuse (D&F)" strategy, reconstructing human body parts independently before fusing them, thereby ensuring robustness against occlusions. We design Human Part Parametric Models (HPPM) that independently reconstruct the mesh from a few shape and global location parameters, without inter-part dependency. A specially designed fusion module then seamlessly integrates the reconstructed parts, even when only a few parts are visible. We harness a large volume of ground truth SMPL data to train our parametric mesh models. To facilitate the training and evaluation of our method, we have established benchmark datasets featuring images of partially visible humans with HPPM annotations. Our experiments, conducted on our benchmark datasets, demonstrate the effectiveness of our D&F method, particularly in scenarios with substantial invisibility, where traditional approaches struggle to maintain reconstruction quality.

In recent years, 2D human pose estimation has made significant progress on public benchmarks. However, many of these approaches face challenges of less applicability in the industrial community due to the large number of parametric quantities and computational overhead. Efficient human pose estimation remains a hurdle, especially for whole-body pose estimation with numerous keypoints. While most current methods for efficient human pose estimation primarily rely on CNNs, we propose the Group-based Token Pruning Transformer (GTPT) that fully harnesses the advantages of the Transformer. GTPT alleviates the computational burden by gradually introducing keypoints in a coarse-to-fine manner. It minimizes the computation overhead while ensuring high performance. Besides, GTPT groups keypoint tokens and prunes visual tokens to improve model performance while reducing redundancy. We propose the Multi-Head Group Attention (MHGA) between different groups to achieve global interaction with little computational overhead. We conducted experiments on COCO and COCO-WholeBody. Compared to other methods, the experimental results show that GTPT can achieve higher performance with less computation, especially in whole-body with numerous keypoints.

Reconstructing hand-held objects from a single RGB image is a challenging task in computer vision. In contrast to prior works that utilize deterministic modeling paradigms, we employ a point cloud denoising diffusion model to account for the probabilistic nature of this problem. In the core, we introduce centroid-fixed dual-stream conditional diffusion for monocular hand-held object reconstruction (D-SCo), tackling two predominant challenges. First, to avoid the object centroid from deviating, we utilize a novel hand-constrained centroid fixing paradigm, enhancing the stability of diffusion and reverse processes and the precision of feature projection. Second, we introduce a dual-stream denoiser to semantically and geometrically model hand-object interactions with a novel unified hand-object semantic embedding, enhancing the reconstruction performance of the hand-occluded region of the object. Experiments on the synthetic ObMan dataset and three real-world datasets HO3D, MOW and DexYCB demonstrate that our approach can surpass all other state-of-the-art methods. Codes will be released.

Head pose estimation (HPE) is crucial for various applications, including human-computer interaction, augmented reality, and driver monitoring. However, traditional RGB-based methods struggle in challenging conditions like sudden movement and extreme lighting. Event cameras, as a neuromorphic sensor, have the advantages of high temporal resolution and high dynamic range, offering a promising solution for HPE. However, the lack of paired event and head pose data hinders the full potential of event-based HPE. To address this, we introduce two large-scale, diverse event-based head pose datasets encompassing 282 sequences across different resolutions and scenarios. Furthermore, we propose the event-based HPE network, featuring two novel modules: the Event Spatial-Temporal Fusion (ESTF) module and the Event Motion Perceptual Attention (EMPA) module. The ESTF module effectively combines spatial and temporal information from the event streams, while the EMPA module captures crucial motion details across the scene using a large receptive field. We also propose a unified loss function to optimize the network using both angle and rotation matrix information. Extensive experiments demonstrate the superior performance of our network on both datasets, showcasing its effectiveness in handling HPE across various challenging scenarios. The datasets and code are available at https://github.com/Jiahui-Yuan-1/EVHPE

We explore the dexterous manipulation transfer problem by designing simulators. The task wishes to transfer human manipulations to dexterous robot hand simulations and is inherently difficult due to its intricate, highly-constrained, and discontinuous dynamics and the need to control a dexterous hand with a DoF to accurately replicate human manipulations. Previous approaches that optimize in high-fidelity black-box simulators or a modified one with relaxed constraints only demonstrate limited capabilities or are restricted by insufficient simulation fidelity. We introduce parameterized quasi-physical simulators and a physics curriculum to overcome these limitations. The key ideas are 1) balancing between fidelity and optimizability of the simulation via a curriculum of parameterized simulators, and 2) solving the problem in each of the simulators from the curriculum, with properties ranging from high task optimizability to high fidelity. We successfully enable a dexterous hand to track complex and diverse manipulations in high-fidelity simulated environments, boosting the success rate by 11\%+ from the best-performed baseline. We include \href{https://quasi-physical-sims.github.io/quasi-physical-sims-for-dex-manip/}{a website} to introduce the work.

sRGB images are now the predominant choice for pre-training visual models in computer vision research, owing to their ease of acquisition and efficient storage. Meanwhile, the advantage of RAW images lies in their rich physical information under variable real-world lighting conditions. For computer vision tasks directly based on camera RAW data, most existing studies adopt methods of integrating image signal processor (ISP) with backend networks, yet often overlook the interaction capabilities between the ISP stages and subsequent networks. Drawing inspiration from ongoing adapter research in NLP and CV areas, we introduce RAW-Adapter, a novel approach aimed at adapting sRGB pre-trained models to camera RAW data. RAW-Adapter comprises input-level adapters that employ learnable ISP stages to adjust RAW inputs, as well as model-level adapters to build connections between ISP stages and subsequent high-level networks. Additionally, RAW-Adapter is a general framework that could be used in various computer vision frameworks. Abundant experiments under different lighting conditions have shown our algorithm's state-of-the-art (SOTA) performance, demonstrating its effectiveness and efficiency across a range of real-world and synthetic datasets. We would release our source code.

This paper proposes a side-view context inpainting strategy (SidePaint) to ease the reasoning of unknown 3D patterns for semantic scene completion. Based on the observation that the learning burden on pattern completion increases with spatial complexity and feature sparsity, the SidePaint strategy is designed to decompose the complex 3D pattern learning into easier 2D context inpainting with dense feature volumes. Specifically, our approach densely lifts image features into 3D volume space with distance-aware projection, and reasons missing patterns in 2D side-view feature maps sliced from feature volumes. With the learning burden relieved by decreasing pattern complexity in 2D space, our SidePaint strategy enables more effective semantic completion than directly learning 3D patterns. Extensive experiments demonstrate the effectiveness of our SidePaint strategy on several challenging semantic scene completion benchmarks.`

We present a novel approach designed to address the complexities posed by challenging, out-of-distribution data in the single-image depth estimation task, including adverse weather conditions and non-Lambertian objects. Starting with images that facilitate depth prediction due to the absence of unfavorable factors, we systematically generate new, user-defined scenes with a comprehensive set of challenges and associated depth information. This is achieved by leveraging cutting-edge conditioned diffusion models, known for their ability to synthesize high-quality image content from textual prompts while preserving the coherence of the 3D structure between generated and source imagery. Subsequent fine-tuning of any monocular depth network, either supervised or self-supervised, is carried out through a self-distillation protocol that takes into account images generated using our strategy and its own depth predictions on simple, unchallenging scenes. Experimental results on benchmarks tailored for our purposes demonstrate the effectiveness and versatility of our proposal\footnote, showing its distinctive ability to simultaneously address adverse weather settings and non-Lambertian objects, and to deliver competitive results with respect to specialized state-of-the-art solutions designed exclusively for each individual challenge.

Monocular depth estimation has greatly improved in the recent years but models predicting metric depth still struggle to generalize across diverse camera poses and datasets. While recent supervised methods mitigate this issue by leveraging ground prior information at inference, their adaptability to self-supervised settings is limited due to the additional challenge of scale recovery. Addressing this gap, we propose in this paper a novel constraint on ground areas designed specifically for the self-supervised paradigm. This mechanism not only allows to accurately recover the scale but also ensures coherence between the depth prediction and the ground prior. Experimental results show that our method surpasses existing scale recovery techniques on the KITTI benchmark and significantly enhances model generalization capabilities. This improvement can be observed by its more robust performance across diverse camera rotations and its adaptability in zero-shot conditions with previously unseen driving datasets such as DDAD.

3D sensing is a fundamental task for Autonomous Vehicles. Its deployment often relies on aligned RGB cameras and LiDAR. Despite meticulous synchronization and calibration, systematic misalignment persists in the LiDAR projected depthmap. This is due to the physical baseline distance between the two sensors. The artifact is often reflected as background LiDAR incorrectly overlayed onto the foreground, such as cars and pedestrian. The KITTI dataset uses stereo cameras as a heuristic solution. However most AV datasets, including nuScenes, Waymo, and DDAD, lack stereo images, making the KITTI solution inapplicable. This work proposes a parameter-free analytical solution to remove the projective artifacts. We construct a binocular vision system between a hypothesized virtual LiDAR camera and the RGB camera. We then remove the projective artifacts by determining the epipolar occlusion with the proposed analytical solution. We show unanimous improvement in the State-of-The-Art (SoTA) monocular depth estimators and 3D object detectors with the artifacts-free depthmaps. Our code and the processed depthmaps of major AV datasets will be publicly available.

Cross-view geolocalization identifies the geographic location of street view images by matching them with a georeferenced satellite database. Significant challenges arise due to the drastic appearance and geometry differences between views. In this paper, we propose a new approach for cross-view image geo-localization, i.e., the Panorama-BEV Co-Retrieval Network. Specifically, by utilizing the ground plane assumption and geometric relations, we convert street view panorama images into the BEV view, reducing the gap between street panoramas and satellite imagery. In the existing retrieval of street view panorama images and satellite images, we introduce BEV and satellite image retrieval branches for collaborative retrieval. By retaining the original street view retrieval branch, we overcome the limited perception range issue of BEV representation. Our network enables comprehensive perception of both the global layout and local details around the street view capture locations. Additionally, we introduce CVGlobal, a global cross-view dataset that is closer to real-world scenarios. This dataset adopts a more realistic setup, with street view directions not aligned with satellite images. CVGlobal also includes cross-regional, cross-temporal, and street view to map retrieval tests, enabling a comprehensive evaluation of algorithm performance. Our method excels in multiple tests on common cross-view datasets such as CVUSA, CVACT, …

Multi-view counting (MVC) methods have shown their superiority over single-view counterparts, particularly in situations characterized by heavy occlusion and severe perspective distortions. However, hand-crafted heuristic features and identical camera layout requirements in conventional MVC methods limit their applicability and scalability. In this work, we propose a concise 3D MVC framework called \textbf{CountFormer} to elevate multi-view image-level features to a scene-level volume representation and estimate the 3D density map based on the volume features. By incorporating a camera encoding strategy, CountFormer successfully embeds camera parameters into the volume query and image-level features, enabling it to handle various camera layouts with significant differences. Furthermore, we introduce a feature lifting module capitalized on the attention mechanism to transform image-level features into a 3D volume representation for each camera view. The multi-view volume aggregation module attentively aggregates various multi-view volumes to create a comprehensive scene-level volume representation, allowing CountFormer to handle images captured by arbitrary dynamic camera layouts. The proposed method performs favorably against the state-of-the-art approaches across various widely used datasets, demonstrating its greater suitability for real-world deployment compared to conventional MVC frameworks.

Online high-definition (HD) map construction is an important and challenging task in autonomous driving. Recently, there has been a growing interest in cost-effective multi-view camera-based methods without relying on other sensors like LiDAR. However, these methods suffer from a lack of explicit depth information, necessitating the use of large models to achieve satisfactory performance. To address this, we employ the Knowledge Distillation (KD) idea for efficient HD map construction for the first time and introduce a novel KD-based approach called MapDistill to transfer knowledge from a high-performance Camera-LiDAR fusion model to a lightweight camera-only model. Specifically, we adopt the teacher-student architecture, i.e., a Camera-LiDAR fusion model as the teacher and a lightweight camera model as the student, and devise a dual BEV transform module to facilitate cross-modal knowledge distillation while maintaining cost-effective camera-only deployment. Additionally, we present a comprehensive distillation scheme encompassing cross-modal relation distillation, dual-level feature distillation, and map head distillation. This approach alleviates knowledge transfer challenges between modalities, enabling the student model to learn improved feature representations for HD map construction. Experimental results on the challenging nuScenes dataset demonstrate the effectiveness of MapDistill, surpassing existing competitors by over 7.7 mAP or 4.5x speedup. The source code will be …

In the realm of autonomous driving, accurate 3D perception is the foundation. However, developing such models relies on extensive human annotations -- a process that is both costly and labor-intensive. To address this challenge from a data representation learning perspective, we introduce SuperFlow, a novel framework designed to harness consecutive LiDAR-camera pairs for establishing spatiotemporal pretraining objectives. SuperFlow stands out by integrating two key designs: 1) a dense-to-sparse consistency regularization, which promotes insensitivity to point cloud density variations during feature learning, and 2) a flow-based contrastive learning module, carefully crafted to extract meaningful temporal cues from readily available sensor calibrations. To further boost learning efficiency, we incorporate a plug-and-play view consistency module that enhances the alignment of the knowledge distilled from camera views. Extensive comparative and ablation studies across 11 heterogeneous LiDAR datasets validate our effectiveness and superiority. Additionally, we observe several interesting emerging properties by scaling up the 2D and 3D backbones during pretraining, shedding light on the future research of 3D foundation models for LiDAR-based perception. Our code is attached to the supplementary material and will be made publicly available.

Accurate object detection in LiDAR point clouds is a key prerequisite of robust and safe autonomous driving and robotics applications. Training the 3D object detectors currently involves the need to manually annotate vasts amounts of training data, which is very time-consuming and costly. As a result, the amount of annotated training data readily available is limited, and moreover these annotated datasets likely do not contain edge-case or otherwise rare instances, simply because the probability of them occurring in such a small dataset is low. In this paper, we propose a method to train 3D object detector without any need for manual annotations, by exploiting existing off-the-shelf vision components and by using the consistency of the world around us. The method can therefore be used to train a 3D detector by only collecting sensor recordings in the real world, which is extremely cheap and allows training using orders of magnitude more data than traditional fully-supervised methods. The method is evaluated on the both KITTI validation and test datasets, where it outperforms all previous weakly-supervised methods and where it narrows the gap when compared to methods using human 3D labels.

Ensuring a reliable perception of vulnerable road users is crucial for safe Autonomous Driving. Radar stands out as an appealing sensor choice due to its resilience in adverse weather, cost-effectiveness, depth sensing capabilities, and established role in adaptive cruise control. Nevertheless, radar's limited angular resolution poses challenges in object recognition, especially in distinguishing targets in close proximity. To tackle this limitation, we present the Camera-Assisted Radar-Based Network (CARB-Net), a novel and efficient framework that merges the angular accuracy of a camera with the robustness and depth sensing capabilities of radar. We integrate camera detection information through a ground plane feed-forward array, entangling it with the early stages of a radar-based detection network. Furthermore, we introduce a unique context learning approach to ensure graceful degradation in situations of poor radar Doppler information or unfavorable camera viewing conditions. Experimental validations on two datasets, along with benchmark comparisons, showcase CARB-Net's superiority, boasting up to a 12% improvement in mAP performance. A series of ablation studies further emphasize the efficacy of the CARB-Net architecture.

Scene flow estimation predicts the 3D motion at each point in successive LiDAR scans. This detailed, point-level, information can help autonomous vehicles to accurately predict and understand dynamic changes in their surroundings. Current state-of-the-art methods require annotated data to train scene flow networks and the expense of labeling inherently limits their scalability. Self-supervised approaches can overcome the above limitations, yet face two principal challenges that hinder optimal performance: point distribution imbalance and disregard for object-level motion constraints. In this paper, we propose SeFlow, a self-supervised method that integrates efficient dynamic classification into a learning-based scene flow pipeline. We demonstrate that classifying static and dynamic points helps design targeted objective functions for different motion patterns. We also emphasize the importance of internal cluster consistency and correct object point association to refine the scene flow estimation, in particular on object details. Our real-time capable method achieves state-of-the-art performance on the self-supervised scene flow task on Argoverse 2 and Waymo datasets. The code is open-sourced at AnonymousforReview along with trained model weights.

In the domain of autonomous driving, the concurrent processing of multiple 3D perception tasks within the same spatiotemporal scene poses a significant challenge, in particular due to the computational inefficiencies and feature competition between tasks when using traditional multi-task learning approaches. This paper addresses these issues by proposing a novel unified representation, RepVF, which harmonizes the representation of various perception tasks such as 3D object detection and 3D lane detection within a single framework. RepVF characterizes the structure of different targets in the scene through a vector field, enabling a single-head, multi-task learning model that significantly reduces computational redundancy and feature competition. Building upon RepVF, we introduce RFTR, a network designed to exploit the inherent connections between different tasks by utilizing a hierarchical structure of queries that implicitly model the relationships both between and within tasks. This approach eliminates the need for task-specific heads and parameters, fundamentally reducing the conflicts inherent in traditional multi-task learning paradigms. We validate our approach by combining labels from the OpenLane dataset with the Waymo Open dataset. Our work presents a significant advancement in the efficiency and effectiveness of multi-task perception in autonomous driving, offering a new perspective on handling multiple 3D perception tasks synchronously …

In autonomous simulation and surveillance, realistic scenarios are crucial for advancing object detection algorithms. Existing aerial datasets suffer from sample class imbalance, especially in larger vehicles like trucks, and unrealistic lighting conditions. This hampers progress in driving behavior analysis and imitation. To address these limitations, we introduce a novel multimodal vehicle surveillance dataset, integrating aerial thermal infrared and sRGB imagery. It contributes: (1) A novel thermal infrared vehicle detection benchmark, ensuring robust object detection in nighttime lighting conditions. (2) Thermal infrared surveillance videos paired with corresponding HD-MAPs for improved multi-vehicle tracking.  (3) Specialized annotations for semi-trailers, precisely documenting their movement trajectories and physical coordinates.TrafficNight significantly advances understanding of larger vehicles in traffic dynamics, serving as a benchmark for enhancing Autopilot systems and traffic surveillance in challenging environments.

The task of separating dynamic objects from static environments using NeRFs has been widely studied in recent years. However, capturing large-scale scenes still poses a challenge due to their complex geometric structures and unconstrained dynamics. Without the help of 3D motion cues, previous methods often require simplified setups with slow camera motion and only a few/single dynamic actors, leading to suboptimal solutions in most urban setups. To overcome such limitations, we present RoDUS, a pipeline for decomposing static and dynamic elements in urban scenes, with thoughtfully separated NeRF models for moving and non-moving components. Our approach utilizes a robust kernel-based initialization coupled with 4D semantic information to selectively guide the learning process. This strategy enables accurate capturing of the dynamics in the scene, resulting in reduced artifacts caused by NeRF on background reconstruction, all by using self-supervision. Notably, experimental evaluations on KITTI-360 and Pandaset datasets demonstrate the effectiveness of our method in decomposing challenging urban scenes into precise static and dynamic components.

Camera-based 3D occupancy prediction has recently garnered increasing attention in outdoor driving scenes. However, research in indoor scenes remains relatively unexplored. The core differences in indoor scenes lie in the complexity of scene scale and the variance in object size. In this paper, we propose a novel method, named ISO, for predicting indoor scene occupancy using monocular images. ISO harnesses the advantages of a pretrained depth model to achieve accurate depth predictions. Subsequently, it employs a Dual Feature Line of Sight Projection(D-FLoSP) module to facilitate the learning of 3D voxel features. Additionally, we introduce a large-scale occupancy benchmark for indoor scenes, titled Occ-ScanNet. With a dataset size 40 times larger than the NYUv2 dataset, it facilitates future scalable research in indoor scene analysis. Experimental results on both NYUv2 and Occ-ScanNet demonstrate that our method achieves state-of-the-art performance. The dataset and code will be made publicly available.

Existing benchmarks for 3D semantic occupancy prediction in autonomous driving are limited by low resolution (up to [512×512×40] with 0.2m voxel size) and inaccurate annotations, hindering the unification of 3D scene understanding through the occupancy representation. Moreover, previous methods can only generate occupancy predictions at 0.4m resolution or lower, requiring post-upsampling to reach their full resolution (0.2m). The root of these limitations lies in the sparsity, noise, and even errors present in the raw data. In this paper, we overcome these challenges by introducing nuCraft, a high-resolution and accurate semantic occupancy dataset derived from nuScenes. nuCraft offers an 8× increase in resolution ([1024 × 1024 × 80] with voxel size of 0.1m) and more precise semantic annotations compared to previous benchmarks. To address the high memory cost of high-resolution occupancy prediction, we propose VQ-Occ, a novel method that encodes occupancy data into a compact latent feature space using a VQ-VAE. This approach simplifies semantic occupancy prediction into feature simulation in the VQ latent space, making it easier and more memory-efficient. Our method enables direct generation of semantic occupancy fields at high resolution without post-upsampling, facilitating a more unified approach to 3D scene understanding. We validate the superior quality of nuCraft …

Predicting vectorized high-definition (HD) map online is useful for autonomous driving, providing detailed geometric and semantic information on the surrounding road environment. In this paper, we introduce Mask2Map, a novel end-to-end online HD map construction method. Our approach identifies semantic components within a scene represented in the bird's eye view (BEV) domain and then generates a precise vectorized map topology based on this information. Mask2Map comprises two main components: an Instance-level Mask Prediction Network (IMPNet) and a Mask-Driven Map Prediction Network (MMPNet). IMPNet generates a mask-aware query capable of producing BEV segmentation masks, while MMPNet accurately constructs vectorized map components, leveraging the semantic geometric information provided by the mask-aware query. For enhancing HD map predictions, we design innovative modules for MMPNet based on outputs from IMPNet. We present a Positional Feature Generator that generates instance-level positional features by utilizing the comprehensive spatial context from semantic components of instance. We also propose a Geometric Feature Extractor which extracts point-level geometric features using sparse key points pooled from the segmentation masks. Furthermore, we present the denoising training strategy for inter-network consistency to boost the performance of map construction. Our evaluation conducted on nuScenes and Argoverse2 benchmarks demonstrates that our Mask2Map achieves a …

We propose CARFF, Conditional Auto-encoded Radiance Field for 3D Scene Forecasting, a method for predicting future 3D scenes given past observations. Our method maps 2D ego-centric images to a distribution over plausible 3D latent scene configurations and predicts the evolution of hypothesized scenes through time. Our latents condition a global Neural Radiance Field (NeRF) to represent a 3D scene model, enabling explainable predictions and straightforward downstream planning. This approach models the world as a POMDP and considers complex scenarios of uncertainty in environmental states and dynamics. Specifically, we employ a two-stage training of Pose-Conditional-VAE and NeRF to learn 3D representations, and auto-regressively predict latent scene representations utilizing a mixture density network. We demonstrate the utility of our method in scenarios using the CARLA driving simulator, where CARFF enables efficient trajectory and contingency planning in complex multi-agent autonomous driving scenarios involving occlusions. Video and code are available at: www.carff.website.

Effectively navigating a dynamic 3D world requires a comprehensive understanding of 3D geometry and motion of surrounding objects and layouts. However, existing methods for perception and planning in autonomous driving primarily rely on a 2D spatial representation, based on a Bird's Eye Perspective of the scene, which are insufficient for modeling motion characteristics and decision-making in real-world 3D settings with occlusion, partial observability, subtle motions, and differing terrains. Motivated by this key insight, we present a novel framework for learning end-to-end autonomous driving based on volumetric representations. Our proposed neural volumetric world modeling approach, NeMo, can be trained in a self-supervised manner for image reconstruction and occupancy prediction tasks, benefiting scalable training and deployment paradigms, such as imitation learning. Specifically, we demonstrate how the higher-fidelity modeling of 3D volumetric representation benefits vision-based motion planning. We further propose a motion flow module to model complex dynamic scenes, enabling additional robust spatial-temporal consistency supervision. Moreover, a temporal attention module is introduced to effectively integrate predicted future volumetric features for the planning task. Our proposed sensorimotor agent achieves state-of-the-art motion planning performance in open-loop evaluation settings in nuScenes, outperforming prior baseline methods by over 18% in L_2 error.

Human trajectory prediction is a practical task of predicting the future positions of pedestrians on the road, which typically covers all temporal ranges from short-term to long-term within a trajectory. However, existing works attempt to address the entire trajectory prediction with a singular, uniform training paradigm, neglecting the distinction between short-term and long-term dynamics in human trajectories. To overcome this limitation, we introduce a novel Progressive Pretext Task learning (PPT) framework, which progressively enhances the model's capacity of capturing short-term dynamics and long-term dependencies for the final entire trajectory prediction. Specifically, we elaborately design three stages of training tasks in the PPT framework. In the first stage, the model learns to comprehend the short-term dynamics through a stepwise next-position prediction task. In the second stage, the model is further enhanced to understand long-term dependencies through a destination prediction task. In the final stage, the model aims to address the entire future trajectory task by taking full advantage of the knowledge from previous stages. To alleviate the knowledge forgetting, we further apply a cross-task knowledge distillation. Additionally, we design a Transformer-based trajectory predictor, which is able to achieve high-efficiency reasoning by integrating a two-step reasoning strategy and a group of parallel …

Planning for the ego vehicle is the ultimate goal of autonomous driving. Although deep learning-based methods have been widely applied to predict future trajectories of other agents in traffic scenes, directly using them to plan for the ego vehicle is often unsatisfactory. This is due to misaligned objectives during training and deployment: a planner that only aims to imitate human driver trajectories is insufficient to accomplish driving tasks well. We argue that existing training processes may not endow models with an understanding of how the physical world evolves. To address this gap, we propose RaSc, which stands for Risk-aware Self-consistent imitation learning. RaSc not only imitates driving trajectories, but also learns the motivations behind human driver behaviors (to be risk-aware) and the consequences of its own actions (by being self-consistent). These two properties stem from our novel prediction branch and training objectives regarding Time-To-Collision (TTC). Moreover, we enable the model to better mine hard samples during training by checking its self-consistency. Our experiments on the large-scale real-world nuPlan dataset demonstrate that RaSc outperforms previous state-of-the-art learning-based methods, in both open-loop and, more importantly, closed-loop settings.

Evaluating the performance of autonomous vehicle planning algorithms necessitates simulating long-tail safety-critical traffic scenarios. However, traditional methods for generating such scenarios often fall short in terms of controllability and realism, and neglect the dynamics of agent interactions. To mitigate these limitations, we introduce Safe-Sim, a novel diffusion-based controllable closed-loop safety-critical simulation framework. Our approach yields two distinct advantages: 1) the generation of realistic long-tail safety-critical scenarios that closely emulate real-world conditions, and 2) enhanced controllability, enabling more comprehensive and interactive evaluations. We develop a novel approach to simulate safety-critical scenarios through an adversarial term in the denoising process, which allows an adversarial agent to challenge a planner with plausible maneuvers, while all agents in the scene exhibit reactive and realistic behaviors. Furthermore, we propose novel guidance objectives and a partial diffusion process that enables a user to control key aspects of the generated scenarios such as the collision type and aggressiveness of the adversarial driver while maintaining the realism of the behavior. We validate our framework empirically using the NuScenes dataset, demonstrating improvements in both realism and controllability. These findings affirm that diffusion models provide a robust and versatile foundation for safety-critical, interactive traffic simulation, extending their utility across the …

“Dual Transparent Liquid” refers to a liquid and its container, both being transparent. Accurately estimating the levels of such a liquid from arbitrary viewpoints is fundamental and crucial, especially in AI-guided autonomous biomedical laboratories for tasks like liquid dispensing, aspiration, and mixing. However, current methods for estimating liquid level focus on scenarios of a single instance captured from a fixed view. We propose a new paradigm for dual transparent liquid level estimation, including a dataset, methods, and practices. The dual transparent liquid dataset, named DTLD, comprises 27,458 images with four object instances captured from multiple views across three biomedical lab scenes. Based on DTLD, we propose an end-to-end learning method for detecting the liquid contact line, followed by an approach to estimate the liquid level. To enhance contact line detection, a color rectification module is proposed to stabilize the color distribution at the local region of the air-liquid interface. Additionally, our method surpasses the current best approach, reducing the mean absolute percentage error by a percentage of 43.4. The dataset and code will be available at https://github.com/dualtransparency/TCLD.

We propose TRAM, a two-stage method to reconstruct a human's global trajectory and motion from in-the-wild videos. TRAM robustifies SLAM to recover the camera motion in the presence of dynamic humans and uses the scene background to derive the motion scale. Using the recovered camera as a metric-scale reference frame, we introduce a video transformer model (VIMO) to regress the kinematic body motion of a human. By composing the two motions, we achieve accurate recovery of 3D humans in the world space, reducing global motion errors by 60% from prior work.

Stereo matching provides depth estimation from binocular images for downstream applications. These applications mostly take video streams as input and require temporally consistent depth maps. However, existing methods mainly focus on the estimation at the single-frame level. This commonly leads to temporally inconsistent results, especially in ill-posed regions. In this paper, we aim to exploit temporal information to improve the temporal consistency and accuracy of stereo matching. To this end, we build a temporally consistent stereo matching network, which includes two stages. In the first stage, we leverage temporal information to obtain a well-initialized disparity. In the second stage, we iteratively refine the disparity based on the temporal initialization. Specifically, we propose a temporal disparity completion module, which completes a semi-dense disparity map transformed from the previous moment. Then, we use a temporal state fusion module to fuse the state of the completion module and the hidden state of the refinement from the previous frame, providing a coherent state for further refinement. Based on this coherent state, we introduce a dual-space refinement module to iteratively refine the initialized result in both the disparity space and the disparity gradient space, improving the estimations in ill-posed regions. Extensive experiments demonstrate that our …

Moving objects are frequently seen in daily life and usually appear blurred in images due to their motion. While general object retrieval is a widely explored area in computer vision, it primarily focuses on sharp and static objects, and retrieval of motion-blurred objects in large image collections remains unexplored. We propose a method for object retrieval in images that are affected by motion blur. The proposed method learns a robust representation capable of matching blurred objects to their deblurred versions and vice versa. To evaluate our approach, we present the first large-scale datasets for blurred object retrieval, featuring images with objects exhibiting varying degrees of blur in various poses and scales. We conducted extensive experiments, showing that our method outperforms state-of-the-art retrieval methods on the new blur-retrieval datasets, which validates the effectiveness of the proposed approach.

The stark contrast in the design philosophy of an event camera makes it particularly ideal for operating under high-speed, high dynamic range and low-light conditions, where standard cameras underperform. Nonetheless, event cameras still suffer from some amount of motion blur, especially under these challenging conditions, in contrary to what most think. This is attributed to the limited bandwidth of the event sensor pixel, which is mostly proportional to the light intensity. Thus, to ensure that event cameras can truly excel in such conditions where it has an edge over standard cameras, it is crucial to account for event motion blur in downstream applications, especially reconstruction. However, none of the recent works on reconstructing Neural Radiance Fields (NeRFs) from events, nor event simulators, have considered the full effects of event motion blur. To this end, we propose, Deblur e-NeRF, a novel method to directly and effectively reconstruct blur-minimal NeRFs from motion-blurred events generated under high-speed motion or low-light conditions. The core component of this work is a physically-accurate pixel bandwidth model proposed to account for event motion blur under arbitrary speed and lighting conditions. We also introduce a novel threshold-normalized total variation loss to improve the regularization of large textureless patches. …

Video deblurring aims to enhance the quality of restored results in motion-blurred videos by effectively gathering information from adjacent video frames to compensate for the insufficient data in a single blurred frame. However, when faced with consecutively severe motion blur situations, frame-based video deblurring methods often fail to find accurate temporal correspondence among neighboring video frames, leading to diminished performance. To address this limitation, we aim to solve the video deblurring task by leveraging an event camera with micro-second temporal resolution. To fully exploit the dense temporal resolution of the event camera, we propose two modules: 1) Intra-frame feature enhancement operates within the exposure time of a single blurred frame, iteratively enhancing cross-modality features in a recurrent manner to better utilize the rich temporal information of events, 2) Inter-frame temporal feature alignment gather valuable long-range temporal information to target frames, aggregating sharp features leveraging the advantages of the events. In addition, we present a novel dataset composed of real-world blurred RGB videos, corresponding sharp videos, and event data. This dataset serves as a valuable resource for evaluating event-guided deblurring methods. We demonstrate that our proposed methods outperform state-of-the-art frame-based and event-based motion deblurring methods through extensive experiments conducted on both …

Long-range imaging inevitably suffers from atmospheric turbulence with severe geometric distortions due to random refraction of light. The further the distance, the more severe the disturbance. Despite existing research has achieved great progress in tackling short-range turbulence, there is less attention paid to long-range turbulence with significant distortions. To address this dilemma and advance the field, we construct a large-scale real long-range atmospheric turbulence dataset (RLR-AT), including 1500 turbulence sequences spanning distances from 1 Km to 13 Km. The advantages of RLR-AT compared to existing ones: turbulence with longer-distances and higher-diversity, scenes with greater-variety and larger-scale. Moreover, most existing work adopts either registration-based or decomposition-based methods to address distortions through one-step mitigation. However, they fail to effectively handle long-range turbulence due to its significant pixel displacements. In this work, we propose a coarse-to-fine framework to handle severe distortions, which cooperates dynamic turbulence and static background priors (CDSP). On the one hand, we discover the pixel motion statistical prior of turbulence, and propose a frequency-aware reference frame for better large-scale distortion registration, greatly reducing the burden of refinement. On the other hand, we take advantage of the static prior of background, and propose a subspace-based low-rank tensor refinement model to eliminate …

Learning rewards from expert videos offers an affordable and effective solution to specify the intended behaviors for reinforcement learning (RL) tasks. In this work, we propose Diffusion Reward, a novel framework that learns rewards from expert videos via conditional video diffusion models for solving complex visual RL problems. Our key insight is that lower generative diversity is exhibited when conditioning diffusion on expert trajectories. Diffusion Reward is accordingly formalized by the negative of conditional entropy that encourages productive exploration of expert behaviors. We show the efficacy of our method over 10 robotic manipulation tasks from MetaWorld and Adroit with visual input and sparse reward. Moreover, Diffusion Reward can even solve unseen tasks successfully and effectively, largely surpassing baseline methods.

Generating realistic human motion is an important task in many computer vision and graphics applications. The rich diversity of human body shapes and sizes significantly influences how people move. However, existing motion models typically ignore these differences and use a normalized, average body size. This leads to a homogenization of motion across human bodies that limits diversity and that may not align with their physical attributes. We propose a novel approach to learn a generative motion model conditioned on body shape. We demonstrate that it is possible to learn such a model from unpaired training data using cycle consistency and intuitive physics and stability constraints that model the correlation between identity and movement. The resulting model produces diverse, physically plausible, dynamically stable human motions that are quantitatively and qualitatively more realistic than the existing state of the art.

The data scarcity problem is a crucial factor that hampers the model performance of IMU-based human motion capture. However, effective data augmentation for IMU-based motion capture is challenging, since it has to capture the physical relations and constraints of human body, while maintaining the data distribution and quality. We propose PoseAugment, a novel pipeline incorporating VAE-based pose generation and physical optimization. Given a pose sequence, the VAE module generates infinite poses with both high fidelity and diversity, while keeping the data distribution. The physical module optimizes poses to satisfy physical constraints with minimal motion restrictions. High-quality IMU data are then synthesized from the augmented poses for training motion capture models. Experiments show that PoseAugment outperforms previous data augmentation and pose generation methods in terms of motion capture accuracy, revealing a strong potential of our method to alleviate the data collection burden for IMU-based motion capture and related tasks driven by human poses.

Human motion generation, a cornerstone technique in animation and video production, has widespread applications in various tasks like text-to-motion and music-to-dance. Previous works focus on developing specialist models tailored for each task without scalability. In this work, we present Large Motion Model (LMM), a motion-centric, multi-modal framework that unifies mainstream motion generation tasks into a generalist model. A unified motion model is appealing since it can leverage a wide range of motion data to achieve broad generalization beyond a single task. However, it is also challenging due to the heterogeneous nature of substantially different motion data and tasks. LMM tackles these challenges from three principled aspects: 1) Data: We consolidate datasets with different modalities, formats and tasks into a comprehensive yet unified motion generation dataset, MotionVerse, comprising 10 tasks, 16 datasets, a total of 320k sequences, and 100 million frames. 2) Architecture: We design an articulated attention mechanism ArtAttention that incorporates body part-aware modeling into Diffusion Transformer backbone. 3) Pre-Training: We propose a novel pre-training strategy for LMM, which employs variable frame rates and masking forms, to better exploit knowledge from diverse training data. Extensive experiments demonstrate that our generalist LMM achieves competitive performance across various standard motion generation tasks …

We introduce the Cross Human Motion Diffusion Model (CrossDiff), a novel approach for generating high-quality human motion based on textual descriptions. Our method integrates 3D and 2D information using a shared transformer network within the training of the diffusion model, unifying motion noise into a single feature space. This enables cross-decoding of features into both 3D and 2D motion representations, regardless of their original dimension. The primary advantage of CrossDiff is its cross-diffusion mechanism, which allows the model to reverse either 2D or 3D noise into clean motion during training. This capability leverages the complementary information in both motion representations, capturing intricate human movement details often missed by models relying solely on 3D information. Consequently, CrossDiff effectively combines the strengths of both representations to generate more realistic motion sequences. In our experiments, our model demonstrates competitive state-of-the-art performance on text-to-motion benchmarks. Moreover, our method consistently provides enhanced motion generation quality, capturing complex full-body movement intricacies. Additionally, with a pretrained model, our approach accommodates using in the wild 2D motion data without 3D motion ground truth during training to generate 3D motion, highlighting its potential for broader applications and efficient use of available data resources.

The field of 3D human motion generation from natural language descriptions, known as Text2Motion, has gained significant attention for its potential application in industries such as film, gaming, and AR/VR. To tackle a key challenge in Text2Motion, the deficiency of 3D human motions and their corresponding textual descriptions, we built a novel large-scale 3D human motion dataset, LaViMo, extracted from in-the-wild web videos and action recognition datasets. LaViMo is approximately 3.3 times larger and encompasses a much broader range of actions than the largest available 3D motion dataset. We then introduce a novel multi-task framework TMT (Text Motion Translator), aimed at generating faithful 3D human motions from natural language descriptions, especially focusing on complicated actions and those not existing in the training set. In contrast to prior works, TMT is uniquely regularized by multiple tasks, including Text2Motion, Motion2Text, Text2Text, and Motion2Motion. This multi-task regularization significantly bolsters the model's robustness and enhances its ability of motion modeling and semantic understanding. Additionally, we devised an augmentation method for the textual descriptions using Large Language Models. This augmentation significantly enhances the model's capability to interpret open-vocabulary descriptions while generating motions. The results demonstrate substantial improvements over existing state-of-the-art methods, particularly in handling diverse …

We present TeSMo, a framework for text-controlled scene-aware motion generation based on denoising diffusion models. Previous text-to-motion methods focus on characters in isolation without considering scenes due to the limited availability of datasets that include motion, text descriptions, and interactive scenes. Our approach begins with pre-training a scene-agnostic text-to-motion diffusion model, emphasizing goal-reaching constraints on large-scale motion-capture datasets. We then enhance this model with a scene-aware component, fine-tuned using data augmented with detailed scene information, including ground plane and object shapes. To facilitate this process, we embed annotated navigation and interaction motions within scenes. The proposed method produces realistic and diverse human-object interactions, such as navigation and sitting, in different scenes with various object shapes, orientations, initial body positions, and poses. Extensive experiments demonstrate that our approach surpasses prior techniques in terms of the plausibility of human-scene interaction, as well as the realism and variety of the generated motions. Code will be released upon publication of this work.

Egocentric gaze anticipation serves as a key building block for the emerging capability of Augmented Reality. Notably, gaze behavior is driven by both visual cues and audio signals during daily activities. Motivated by this observation, we introduce the first model that leverages both the video and audio modalities for egocentric gaze anticipation. Specifically, we propose a Contrastive Spatial-Temporal Separable (CSTS) fusion approach that adopts two modules to separately capture audio-visual correlations in spatial and temporal dimensions, and applies a contrastive loss on the re-weighted audio-visual features from fusion modules for representation learning. We conduct extensive ablation studies and thorough analysis using two egocentric video datasets: Ego4D and Aria, to validate our model design. We demonstrate the audio improves the performance by +2.5% and +2.4% on the two datasets. Our model also outperforms the prior state-of-the-art methods by at least +1.9% and +1.6%. Moreover, we provide visualizations to show the gaze anticipation results and provide additional insights into audio-visual representation learning.

Video-to-audio (V2A) generation leverages visual-only video features to render plausible sounds that match the scene. Importantly, the generated sound onsets should match the visual actions that are aligned with them, otherwise unnatural synchronization artifacts arise. Recent works have explored the progression of conditioning sound generators on still images and then video features, focusing on quality and semantic matching while ignoring synchronization, or by sacrificing some amount of quality to focus on improving synchronization only. In this work, we propose a V2A generative model, named MaskVAT, that interconnects a full-band high-quality general audio codec with a sequence-to-sequence masked generative model. This combination allows modeling both high audio quality, semantic matching, and temporal synchronicity at the same time. Our results show that, by combining a high-quality codec with the proper pre-trained visual features and a sequence-to-sequence parallel structure, we are able to yield highly synchronized results on one hand, whilst being competitive with the state of the art of non-codec generative audio models.

In this paper, we introduce PoseCrafter, a one-shot method for personalized video generation following the control of flexible poses. Built upon Stable Diffusion and ControlNet, we carefully design an inference process to produce high-quality videos without the corresponding ground-truth frames. First, we select an appropriate reference frame from the training video and invert it to initialize all latent variables for generation. Then, we insert the corresponding training pose into the target pose sequences to enhance faithfulness through a trained temporal attention module. Furthermore, to alleviate the face and hand degradation resulting from discrepancies between poses of training videos and inference poses, we implement simple latent editing through an affine transformation matrix involving facial and hand landmarks. Extensive experiments on several datasets demonstrate that PoseCrafter achieves superior results to baselines pre-trained on a vast collection of videos under 8 commonly used metrics. Besides, PoseCrafter can follow poses from different individuals or artificial edits and simultaneously retain the human identity in an open-domain training video.

While recent years have witnessed great progress on using diffusion models for video generation, most of them are simple extensions of image generation frameworks, which fail to explicitly consider one of the key differences between videos and images, i.e., motion. In this paper, we propose a novel motion-aware video generation (MoVideo) framework that takes motion into consideration from two aspects: video depth and optical flow. The former regulates motion by per-frame object distances and spatial layouts, while the later describes motion by cross-frame correspondences that help in preserving fine details and improving temporal consistency. More specifically, given a key frame that exists or generated from text prompts, we first design a diffusion model with spatio-temporal modules to generate the video depth and the corresponding optical flows. Then, the video is generated in the latent space by another spatio-temporal diffusion model under the guidance of depth, optical flow-based warped latent video and the calculated occlusion mask. Lastly, we use optical flows again to align and refine different frames for better video decoding from the latent space to the pixel space. In experiments, MoVideo achieves state-of-the-art results in both text-to-video and image-to-video generation, showing promising prompt consistency, frame consistency and visual quality.

Though video generation has witnessed rapid progress, the inference results of existing models still exhibit unsatisfactory temporal consistency and unnatural dynamics. In this paper, we delve deep into the noise initialization of video diffusion models, and discover an implicit training-inference gap that attributes to the inference quality drop. Our key findings are: 1) the spatial-temporal frequency distribution of the initial latent's signal-to-noise ratio (SNR) at inference is intrinsically different from training, and 2) the denoising process is significantly influenced by the low-frequency component of the initial noise. Motivated by these observations, we propose a concise yet effective inference sampling strategy, FreeInit, which significantly improves temporal consistency of videos generated by diffusion models. Through iteratively refining the spatial-temporal low-frequency component of the initial latent during inference, FreeInit is able to compensate the initialization gap between training and inference, thus effectively improving the subject appearance and temporal consistency of generation results. Extensive experiments demonstrate that FreeInit consistently enhances the generation results of various text-to-video generation models without additional training.

Text-driven diffusion-based video editing presents a unique challenge not encountered in image editing literature: establishing real-world motion. Unlike existing video editing approaches, here we focus on score distillation sampling to circumvent the standard reverse diffusion process and initiate optimization from videos that already exhibit natural motion. Our analysis reveals that while video score distillation can effectively introduce new content indicated by target text, it can also cause significant structure and motion deviation. To counteract this, we propose to match space-time self-similarities of the original video and the edited video during the score distillation. Thanks to the use of score distillation, our approach is model-agnostic, which can be applied for both cascaded and non-cascaded video diffusion frameworks. Through extensive comparisons with leading methods, our approach demonstrates its superiority in altering appearances while accurately preserving the original structure and motion. Codes and data will be released upon publication. (Project page: https://anony12anony34.github.io/ )

We introduce Videoshop, a training-free video editing algorithm for localized semantic edits. Videoshop allows users to use any editing software, including Photoshop and generative inpainting, to modify the first frame; it automatically propagates those changes, with semantic, spatial, and temporally consistent motion, to the remaining frames. Unlike existing methods that enable edits only through imprecise textual instructions, Videoshop allows users to add or remove objects, semantically change objects, insert stock photos into videos, etc. with fine-grained control over locations and appearance. We achieve this through image-based video editing by inverting latents with noise extrapolation, from which we generate videos conditioned on the edited image. Videoshop produces higher quality edits against 6 baselines on 2 editing benchmarks using 10 evaluation metrics.

Recent advancements in text-guided diffusion models have unlocked powerful image manipulation capabilities. However, applying these methods to real images necessitates the inversion of the images into the domain of the pretrained diffusion model. Achieving faithful inversion remains a challenge, particularly for more recent models trained to generate images with a small number of denoising steps. In this work, we introduce an inversion method with a high quality-to-operation ratio, enhancing reconstruction accuracy without increasing the number of operations. Building on reversing the diffusion sampling process, our method employs an iterative renoising mechanism at each inversion sampling step. This mechanism refines the approximation of a predicted point along the forward diffusion trajectory, by iteratively applying the pretrained diffusion model, and averaging these predictions. We evaluate the performance of our ReNoise technique using various sampling algorithms and models, including recent accelerated diffusion models. Through comprehensive evaluations and comparisons, we show its effectiveness in terms of both accuracy and speed. Furthermore, we confirm that our method preserves editability by demonstrating text-driven image editing on real images.

The electronic writing tools, while enhancing convenience, sacrifice the readability and efficiency of handwritten content. Balancing high efficiency with readable handwriting poses a challenging research task. In this paper, we propose a method sequence-based models to beautify user handwritten traces. Unlike most existing methods that treat Chinese handwriting as images and cannot reflect the human writing process, we capture individual writing characteristics from a small amount of user handwriting trajectories and beautify the user's traces by mimicking their writing style and process. We fully consider the style of radicals and components between the content and reference glyphs, assigning appropriate fine-grained styles to strokes in the content glyphs through a cross-attention mechanism module. Additionally, we find that many style features contribute minimally to the final stylized results. Therefore, we decompose the style features into the Cartesian product of single-dimensional variable sets, effectively removing redundant features with limited impact on the stylization effect while preserving key style information. Qualitative and quantitative experiments both demonstrate the superiority of our approach.

Existing handwritten text generation methods often require more than ten handwriting samples as style references. However, in practical applications, users tend to prefer a handwriting generation model that operates with just a single reference sample for its convenience and efficiency. This approach, known as ''one-shot generation'', significantly simplifies the process but poses a significant challenge due to the difficulty of accurately capturing a writer's style from a single sample, especially when extracting fine details from the characters' edges amidst sparse foreground and undesired background noise. To address this problem, we propose a One-shot Diffusion Mimicker (One-DM) to generate handwritten text that can mimic any calligraphic style with only one reference sample. Inspired by the fact that high-frequency information of the individual sample often contains distinct style patterns (e.g., character slant and letter joining), we develop a novel style-enhanced module to improve the style extraction by incorporating high-frequency components from a single sample. We then fuse the style features with the text content as a merged condition for guiding the diffusion model to produce high-quality handwritten text images. Extensive experiments demonstrate that our method can successfully generate handwriting scripts with just one sample reference in multiple languages, even outperforming previous methods …

Generative models are now widely used by graphic designers and artists. Prior works have shown that these models tend to remember and often replicate content from the training data during generation. Hence as their proliferation increase, it has become important to perform a database search to determine whether the properties of the image are attributable to specific training data, every time before a generated image is used for professional purposes. Existing tools for this purpose focus largely on retrieving images of similar semantic content. Meanwhile, many artists are concerned with the extent of style replication in text-to-image models. We present a framework to understand and extract style descriptors from images. Our framework comprises a new dataset curated using the insight that style is a subjective property of an image capturing complex yet meaningful interactions of factors including but not limited to colors, textures, shapes, etc. We also propose a method to extract style descriptors that can be used to attribute style of a generated image to the images used in the training dataset of a text-to-image model. We show promising results in various style retrieval tasks. We also quantitatively and qualitatively analyze style attribution and matching in the Stable Diffusion …

Enabling machines to understand structured visuals like slides and user interfaces is essential for making them accessible to people with disabilities. However, achieving such understanding computationally has required manual data collection and annotation, which is time-consuming and labor-intensive. To overcome this challenge, we present a method to generate synthetic, structured visuals with target labels using code generation. Our method allows people to create datasets with built-in labels and train models with a small number of human-annotated examples. We demonstrate performance improvements in three tasks for understanding slides and UIs: recognizing visual elements, describing visual content, and classifying visual content types.

This paper propels creative control in generative visual AI by allowing users to ``select''. Departing from traditional text or sketch-based methods, we for the first time allow users to choose visual part concepts for their creative endeavors. The outcome is fine-grained generation that precisely captures selected visual part concepts, ensuring a holistically faithful and plausible result. To achieve this, we first parse objects into parts through unsupervised feature clustering. Then, we encode parts into text tokens and introduce an entropy-based normalized attention loss that operates on them. This loss design enables our model to learn generic prior topology knowledge about object's part composition, and further generalize to novel part compositions to ensure the generation looks holistically faithful. Lastly, we employ a bottleneck encoder to project the part tokens, this not only enhances fidelity but also accelerates learning, by leveraging shared knowledge and facilitating information exchange among instances. Visual results in the paper and supplementary material showcase the compelling power of PartCraft in crafting highly customized, innovative creations.

We introduce DreamDrone, a novel zero-shot and training-free pipeline for generating unbounded flythrough scenes from textual prompts. Different from other methods that focus on warping images frame by frame, we advocate explicitly warping the intermediate latent code of the pre-trained text-to-image diffusion model for high-quality image generation and generalization ability. To further enhance the fidelity of the generated images, we also propose a feature-correspondence-guidance diffusion process and a high-pass filtering strategy to promote geometric consistency and high-frequency detail consistency, respectively. Extensive experiments reveal that DreamDrone significantly surpasses existing methods, delivering highly authentic scene generation with exceptional visual quality, without training or fine-tuning on datasets or reconstructing 3D point clouds in advance.

Accurate text segmentation results are crucial for text-related generative tasks, such as text image generation, text editing, text removal, and text style transfer. Recently, some scene text segmentation methods have made significant progress in segmenting regular text. However, these methods perform poorly in scenarios containing artistic text. Therefore, this paper focuses on the more challenging task of artistic text segmentation and constructs a real artistic text segmentation dataset. One challenge of the task is that the local stroke shapes of artistic text are changeable with diversity and complexity. We propose a decoder with the layer-wise momentum query to prevent the model from ignoring stroke regions of special shapes. Another challenge is the complexity of the global topological structure. We further design a skeleton-assisted head to guide the model to focus on the global structure. Additionally, to enhance the generalization performance of the text segmentation model, we propose a strategy for training data synthesis, based on the large multi-modal model and the diffusion model. Experimental results show that our proposed method and synthetic dataset can significantly enhance the performance of artistic text segmentation and achieve state-of-the-art results on other public datasets. We will release our datasets and source code.

General text-to-image models bring revolutionary innovation to the fields of arts, design, and media. However, when applied to garment generation, even the state-of-the-art text-to-image models suffer from fine-grained semantic misalignment, particularly concerning the quantity, position, and interrelations of garment components. Addressing this, we propose GarmentAligner, a text-to-garment diffusion model trained with retrieval-augmented multi-level corrections. To achieve semantic alignment at the component level, we introduce an automatic component extraction pipeline to obtain spatial and quantitative information of garment components from corresponding images and captions. Subsequently, to exploit component relationships within the garment images, we construct retrieval subsets for each garment by retrieval augmentation based on component-level similarity ranking and conduct contrastive learning to enhance the model perception of components from positive and negative samples. To further enhance the alignment of components across semantic, spatial, and quantitative granularities, we propose the utilization of multi-level correction losses that leverage detailed component information. The experimental findings demonstrate that GarmentAligner achieves superior fidelity and fine-grained semantic alignment when compared to existing competitors.

In this paper, we introduce PixArt-Sigma, a Diffusion Transformer model~(DiT) capable of directly generating images at 4K resolution. PixArt-Sigma represents a significant advancement over its predecessor, PixArt-Alpha, offering images of markedly higher fidelity and improved alignment with text prompts. A key feature of PixArt-Sigma is its training efficiency. Leveraging the foundational pre-training of PixArt-Alpha, it evolves from the weaker' baseline to astronger' model via incorporating higher quality data, a process we term ``weak-to-strong training''. The advancements in PixArt-Sigma are twofold: (1) High-Quality Training Data: PixArt-Sigma incorporates superior-quality image data, paired with more precise and detailed image captions. (2) Efficient Token Compression: we propose a novel attention module within the DiT framework that compresses both keys and values, significantly improving efficiency and facilitating ultra-high-resolution image generation. Thanks to these improvements, PixArt-Sigma achieves superior image quality and user prompt adherence capabilities with significantly smaller model size~(0.6B parameters) than existing text-to-image diffusion models, such as SDXL~(2.6B parameters) and SD Cascade~(5.1B parameters). Moreover, PixArt-Sigma's capability to generate 4K images supports the creation of high-resolution posters and wallpapers, efficiently bolstering the production of high-quality visual content in industries such as film and gaming.

Subject-driven generation for text-to-image diffusion models aims to encode and inverse specific textual prompts in order to generate personalized images with particular content. Previous studies successfully achieved such a goal by using an optimization-based textual inversion or direct-regression-based concept encoding strategy. However, there are still challenges on how to realize fast and effective prompt inversion while guaranteeing the generalization of the original diffusion models. Motivated by the advantages of both optimization-based and direct-regression-based methods, in this study we proposed a novel hybrid prompt inversion framework called ~\name~ to achieve efficient subject-driven generation of text-to-image diffusion models. In detail, we address the limitations caused by the current optimization-based and direct-regression-based methods by designing a novel hybrid prompt inversion framework and combining it with a mask-guided multi-word text encoding module to enable a fast and robust prompt inversion, additionally, we import a hybrid textual feature fusion module to enhance the representation of the textual feature during learning. As a result, our framework manages to inverse arbitrary visual concepts to a pre-trained diffusion model in an effective and fast way even learning from a single image, and maintaining the general generation ability of the original model. The extensive experiments reveal the effectiveness of …

With the growing popularity of text-to-image generative models, there has been increasing focus on understanding their risks and biases. Recent work has found that state-of-the-art models struggle to depict everyday objects with the true diversity of the real world and have notable gaps between geographic regions. In this work, we aim to increase the diversity of generated images of common objects such that per-region variations are representative of the real world. We introduce an inference-time intervention, contextualized Vendi Score Guidance (c-VSG), that guides the backwards steps of latent diffusion models to increase the diversity of a sample as compared to a ``memory bank'' of previously generated images while constraining the amount of variation within that of an exemplar set of real-world contextualizing images. We evaluate c-VSG with two geographically representative datasets and find that it substantially increases the diversity of generated images, both for the worst-performing regions and on average, while simultaneously maintaining or improving image quality and consistency. Additionally, qualitative analyses reveal that diversity of generated images is significantly improved, including along the lines of reductive region portrayals present in the original model. We hope that this work is a step towards text-to-image generative models that reflect the true …

Drawing on recent advancements in diffusion models for text-to-image generation, identity-preserved personalization has made significant progress in accurately capturing specific identities with just a single reference image. However, existing methods primarily integrate reference images within the text embedding space, leading to a complex entanglement of image and text information, which poses challenges for preserving both identity fidelity and semantic consistency. To tackle this challenge, we propose Infinite-ID, an ID-semantics decoupling paradigm for identity-preserved personalization. Specifically, we introduce identity-enhanced training, incorporating an additional image cross-attention module to capture sufficient ID information while deactivating the original text cross-attention module of the diffusion model. This ensures that the image stream faithfully represents the identity provided by the reference image while mitigating interference from textual input. Additionally, we introduce a feature interaction mechanism that combines a mixed attention module with an AdaIN-mean operation to seamlessly merge the two streams. This mechanism not only enhances the fidelity of identity and semantic consistency but also enables convenient control over the styles of the generated images. Extensive experimental results on both raw photo generation and style image generation demonstrate the superior performance of our proposed method.

Recent advancements in text-to-image diffusion models have demonstrated their remarkable capability to generate high-quality images from textual prompts. However, increasing research indicates that these models memorize and replicate images from their training data, raising tremendous concerns about potential copyright infringement and privacy risks. In our study, we provide a novel perspective to understand this memorization phenomenon by examining its relationship with cross-attention mechanisms. We reveal that during memorization, the cross-attention tends to focus disproportionately on the embeddings of specific tokens. The diffusion model is overfitted to these token embeddings, memorizing corresponding training images. To elucidate this phenomenon, we further identify and discuss various intrinsic findings of cross-attention that contribute to memorization. Building on these insights, we introduce an innovative approach to detect and mitigate memorization in diffusion models. The advantage of our proposed method is that it will not compromise the speed of either the training or the inference processes in these models while preserving the quality of generated images.

Concept erasure in text-to-image diffusion models aims to disable pre-trained diffusion models from generating images related to a target concept. To perform reliable concept erasure, the properties of robustness and locality are desirable. The former refrains the model from producing images associated with the target concept for any paraphrased or learned prompts, while the latter preserves its ability in generating images with non-target concepts. In this paper, we propose Reliable Concept Erasing via Lightweight Erasers (Receler). It learns a lightweight Eraser to perform concept erasing while satisfying the above desirable properties by proposed concept-localized regularization and adversarial prompt learning schemes. Comprehensive experiments with various concepts verify the superiority of Receler over previous methods. Our code will be available upon acceptance.

Previous techniques for face reenactment and swapping predominantly rely on GAN frameworks. However, recent research has shifted its focus towards leveraging large diffusion models for these tasks, owing to their superior generation capabilities. Nonetheless, training these models incurs significant computational costs, and the results have not yet attained satisfactory performance levels. To address this issue, we introduce Face-Adapter, an efficient and effective adapter designed for high-precision and high-fidelity face editing by pretrained diffusion models, which contains: 1) Spatial Condition Generator provides precise landmarks and background; 2) Plug-and-play Identity Encoder transfers face embeddings to the text space by a transformer decoder. 3) Attribute Controller integrates spatial condition and detailed attributes. Face-Adapter achieves comparable or even superior performance in terms of motion control precision, ID retention capability, and generation quality compared to fully fine-tuned models in face reenactment/swapping tasks. Additionally, Face-Adapter seamlessly integrates with popular pretrained diffusion models such as StableDiffusion. Full codes will be available.

We propose a permutation-based explanation method for image classifiers. Current image model explanations like activation maps are limited to instance-based explanations in the pixel space, making it difficult to understand global model behavior. Permutation based explanations for tabular data classifiers measure feature importance by comparing original model performance to model performance on data after permuting a feature. We propose an explanation method for image-based models that permutes interpretable concepts across dataset images. Given a dataset of images labeled with specific concepts like captions, we permute a concept across examples and then generate images via a text-conditioned diffusion model. Model importance is then given by the change in classifier performance relative to unpermuted data. When applied to a set of concepts, the method generates a ranking of concept importance. We show that this approach recovers underlying model feature importance on synthetic and real-world image classification tasks.

Diffusion models have proven to be state-of-the-art methods for generative tasks. These models involve training a U-Net to iteratively predict and remove noise, and the resulting model can synthesize high-fidelity, diverse, novel images. However, text-free diffusion models have typically not been explored for discriminative tasks. In this work, we take a pre-trained unconditional diffusion model and analyze its features post hoc. We find that the intermediate feature maps of the pre-trained U-Net are diverse and have hidden discriminative representation properties. To unleash the potential of these latent properties of diffusion models, we present novel aggregation schemes. Firstly, we propose a novel attention mechanism for pooling feature maps and further leverage this mechanism as DifFormer, a transformer feature fusion of different diffusion U-Net blocks and noise steps. Next, we also develop DifFeed, a novel feedback mechanism tailored to diffusion. We find that diffusion models are better than GANs, and, with our fusion and feedback mechanisms, can compete with state-of-the-art representation learning methods for discriminative tasks -- image classification with full and semi-supervision, transfer for fine-grained classification, object detection, and semantic segmentation.

While text-to-image diffusion models have been shown to achieve state-of-the-art results in image synthesis, they have yet to prove their effectiveness in downstream applications. Previous work has proposed to generate data for image classifier training given limited real data access. However, these methods struggle to generate in-distribution images or depict fine-grained features, thereby hindering the generalization of classification models trained on synthetic datasets. We propose DataDream, a framework for synthesizing classification datasets that more faithfully represents the real data distribution when guided by few-shot examples of the target classes. DataDream fine-tunes LoRA weights for the image generation model on the few real images before generating the training data using the adapted model. We then fine-tune LoRA weights for CLIP using the synthetic data to improve downstream image classification over previous approaches on a large variety of datasets. We demonstrate the efficacy of DataDream through extensive experiments, surpassing state-of-the-art classification accuracy with few-shot data across 9 out of 10 datasets. Additionally, we provide insights into the impact of various factors, such as the number of real-shot and generated images as well as the fine-tuning compute on model performance.

Due to the difficulty and labor-consuming nature of getting highly accurate or matting annotations, there only exists a limited amount of highly accurate labels available to the public. To tackle this challenge, we propose a DiffuMatting which inherits the strong Everything generation ability of diffusion and endows the power of `matting anything'. Our DiffuMatting can 1). act as an anything matting factory with high accurate annotations 2). be well-compatible with community LoRAs or various conditional control approaches to achieve the community-friendly art design and controllable generation. Specifically, inspired by green-screen-matting, we aim to teach the diffusion model to paint on a fixed green screen canvas. To this end, a large-scale green-screen dataset (Green100K) is collected as a training dataset for DiffuMatting. Secondly, a green background control loss is proposed to keep the drawing board as a pure green color to distinguish the foreground and background. To ensure the synthesized object has more edge details, a detailed-enhancement of transition boundary loss is proposed as a guideline to generate objects with more complicated edge structures. Aiming to simultaneously generate the object and its matting annotation, we build a matting head to make a green-color removal in the latent space of the VAE …

We propose ZeST, a method for zero-shot material transfer to an object in the input image given a material exemplar image. ZeST leverages existing diffusion adapters to extract implicit material representation from the exemplar image. This representation is used to transfer the material using pre-trained inpainting diffusion model on the object in the input image using depth estimates as geometry cue and grayscale object shading as illumination cues. The method works on real images without any training resulting a zero-shot approach. Both qualitative and quantitative results on real and synthetic datasets demonstrate that ZeST outputs photorealistic images with transferred materials. We also show the application of ZeST to perform multiple edits and robust material assignment under different illuminations.

We offer a novel approach to image composition, which integrates multiple input images into a single, coherent image. Rather than concentrating on specific use cases such as appearance editing (image harmonization) or semantic editing (semantic image composition), we showcase the potential of utilizing the powerful generative prior inherent in large-scale pre-trained diffusion models to accomplish generic image composition applicable to both scenarios. We observe that the pre-trained diffusion models automatically identify simple copy-paste boundary areas as low-density regions during denoising. Building on this insight, we propose to optimize the composed image towards high-density regions guided by the diffusion prior. In addition, we introduce a novel mask-guided loss to further enable flexible semantic image composition. Extensive experiments validate the superiority of our approach in achieving generic zero-shot image composition. Additionally, our approach shows promising potential in various tasks, such as object removal and multi-concept customization.

Existing imaging systems support wide-gamut images like ProPhoto RGB, but most images are typically encoded in a narrower gamut space (e.g., sRGB). To this end, these images can be enhanced by learning to recover the original color values beyond the sRGB gamut, or out-of-gamut values. Current methods incorporate the metadata from the target wide-gamut images to expand the gamut, while preventing distortion of in-gamut values. However, this metadata is hard to obtain in real-world scenarios. In this paper, we propose a novel method that requires no metadata. We formulate gamut expansion as a ``root-finding" problem and learn an equilibrium transformation via a neural network. Specifically, our method defines a dynamic system that keeps in-gamut values stable to prevent color distortion and updates out-of-gamut values recurrently. Therefore, we employ an implicit recurrent mechanism to iteratively extract features, which can effectively mitigate the vanishing gradient problem, and reduce the GPU memory consumption to O(1) complexity. Experiments demonstrate the effectiveness and efficiency of our model, in terms of gamut expansion and color restoration, outperforming state-of-the-art models by 0.40dB, in terms of PSNR, with a size of 40K parameters only.

Diffusion models have marked a significant breakthrough in the synthesis of semantically coherent images. However, their extensive noise estimation networks and the iterative generation process limit their wider application, particularly on resource-constrained platforms like mobile devices. Existing post-training quantization (PTQ) methods have managed to compress diffusion models to low precision. Nevertheless, due to the iterative nature of diffusion models, quantization errors tend to accumulate throughout the generation process. This accumulation of error becomes particularly problematic in low-precision scenarios, leading to significant distortions in the generated images. We attribute this accumulation issue to two main causes: error propagation and exposure bias. To address these problems, we propose a timestep-aware correction method for quantized diffusion model, which dynamically corrects the quantization error. By leveraging the proposed method in low-precision diffusion models, substantial enhancement of output quality could be achieved with only negligible computation overhead. Extensive experiments underscore our method's effectiveness and generalizability. By employing the proposed correction strategy, we achieve state-of-the-art (SOTA) results on low-precision models.

In this paper, we propose a novel generative model that utilizes a conditional Energy-Based Model (EBM) for enhancing Variational Autoencoder (VAE), termed Energy-Calibrated VAE (EC-VAE). Specifically, VAEs often suffer from blurry generated samples due to the lack of a tailored training on the samples generated in the generative direction. On the other hand, EBMs can generate high-quality samples but require expensive Markov Chain Monte Carlo (MCMC) sampling. To address these issues, we introduce a conditional EBM for calibrating the generative direction of VAE during training, without requiring it for the generation at test time. In particular, we train EC-VAE upon both the input data and the calibrated samples with adaptive weight to enhance efficacy while avoiding MCMC sampling at test time. Furthermore, we extend the calibration idea of EC-VAE to variational learning and normalizing flows, and apply EC-VAE to an additional application of zero-shot image restoration via neural transport prior and range-null theory. We evaluate the proposed method with two applications, including image generation and zero-shot image restoration, and the experimental results show that our method achieves competitive performance over single-step non-adversarial generation.

In order to improve the quality of synthesized videos, currently, one predominant method involves retraining an expert diffusion model and then implementing a noising-denoising process for refinement. Despite the significant training costs, maintaining content consistency between the original and enhanced videos remains a major challenge. To tackle this challenge, we propose a novel formulation that considers both visual quality and content consistency. Content consistency is ensured by a proposed loss fuction that maintains the structure of the input, while visual quality is improved by utilizing the denoising process of pretrained diffusion models. To address the formulated optimization problem, we have developed a plug-and-play noise optimization strategy, which we refer to as \textbf{Noise Calibration}. By simply refining the initial random noise through a few iterations, the original video content can be largely preserved, and the enhancement effect is significantly improved. Extensive experiments have demonstrated the effectiveness of the proposed method.

Existing methods attempt to improve models' generalization ability on real-world hazy images by exploring well-designed training schemes (e.g., CycleGAN, prior loss). However, most of them need very complicated training procedures to achieve satisfactory results. For the first time, we present a novel pipeline called Prompt-based Test-Time Dehazing (PTTD) to help generate visually pleasing results of real-captured hazy images during the inference phase. We experimentally observe that given a dehazing model trained on synthetic data, fine-tuning the statistics (i.e., mean and standard deviation) of encoding features is able to narrow the domain gap, boosting the performance of real image dehazing. Accordingly, we first apply a prompt generation module (PGM) to generate a visual prompt, which is the reference of appropriate statistical perturbations for mean and standard deviation. Then, we employ a feature adaptation module (FAM) into the existing dehazing models for adjusting the original statistics with the guidance of the generated prompt. PTTD is model-agnostic and can be equipped with various state-of-the-art dehazing models trained on synthetic hazy-clean pairs to tackle the real image dehazing task. Extensive experimental results demonstrate that our PTTD is effective, achieving superior performance against state-of-the-art dehazing methods in real-world scenarios.

In recent years, self-supervised denoising methods have gained significant success and become critically important in the field of image restoration. Among them, the blind spot network based methods are the most typical type and have attracted the attentions of a large number of researchers. Although the introduction of blind spot operations can prevent identity mapping from noise to noise, it imposes stringent requirements on the receptive fields in the network design, thereby limiting overall performance. To address this challenge, we propose a single mask scheme for self-supervised denoising training, which eliminates the need for blind spot operation and thereby removes constraints on the network structure design. Furthermore, to achieve denoising across entire image during inference, we propose a multi-mask scheme. Our method, featuring the asymmetric mask scheme in training and inference, achieves state-of-the-art performance on existing real noisy image datasets. All the source code will be made available to the public.

Traditional single-task image restoration methods excel in handling specific degradation types but struggle with multiple degradations. To address this limitation, we propose Grouped Restoration with Image Degradation Similarity (GRIDS), a novel approach that harmonizes the competing objectives inherent in multiple-degradation restoration. We first introduce a quantitative method for assessing relationships between image degradations using statistical modeling of deep degradation representations. This analysis facilitates the strategic grouping of similar tasks, enhancing both the efficiency and effectiveness of the restoration process. Based on the degradation similarity, GRIDS divides restoration tasks into optimal groups, where tasks within the same group are highly correlated. For instance, GRIDS effectively groups 11 degradation types into 4 cohesive groups. Trained models within each group show significant improvements, with an average improvement of 0.09dB over single-task upper bound models and 2.24dB over the mix-training baseline model. GRIDS incorporates an adaptive model selection mechanism for inference, automatically selecting the appropriate grouped-training model based on the input degradation. This mechanism is particularly useful for real-world scenarios with unknown degradations as it does not rely on explicit degradation classification modules. Furthermore, our method can predict model generalization ability without the need for network inference, providing valuable insights for practitioners.

Scale arbitrary super-resolution based on implicit image function gains increasing popularity since it can better represent the visual world in a continuous manner. However, existing scale arbitrary works are trained and evaluated on simulated datasets, where low-resolution images are generated from their ground truths by the simplest bicubic downsampling. These models exhibit limited generalization to real-world scenarios due to the greater complexity of real-world degradations. To address this issue, we build a RealArbiSR dataset, a new real-world super resolution benchmark with both integer and non-integer scaling factors for the training and evaluation of real-world scale arbitrary super-resolution. Moreover, we propose a Dual-level Deformable Implicit Representation (DDIR) to solve real-world scale arbitrary super-resolution. Specifically, we design the appearance embedding and deformation field to handle both image-level and pixel-level deformations caused by real-world degradations. The appearance embedding models the characteristics of low-resolution inputs to deal with photometric variations at different scales, and the pixel-based deformation field learns RGB differences which result from the deviations between the real-world and simulated degradations at arbitrary coordinates. Extensive experiments show our trained model achieves state-of-the-art performance on the RealArbiSR and RealSR benchmarks for real-world scale arbitrary super resolution. The dataset and code are available at https://github.com/nonozhizhiovo/RealArbiSR.

Recent Blind Image Super-Resolution (BSR) methods have shown proficiency in general images. However, we find that the efficacy of recent methods obviously diminishes when employed on image data with blur, while images with intentional blur constitute a substantial proportion of general data. To further investigate and address this issue, we developed a new super-resolution dataset specifically tailored for blur images, named the Real-world Blur-kept Super-Resolution (ReBlurSR) dataset, which consists of nearly 3000 defocus and motion blur image samples with diverse blur sizes and varying blur intensities. Furthermore, we propose a new BSR framework for blur images called Perceptual-Blur-adaptive Super-Resolution (PBaSR), which comprises two main modules: the Cross Disentanglement Module (CDM) and the Cross Fusion Module (CFM). The CDM utilizes a dual-branch parallelism to isolate conflicting blur and general data during optimization. The CFM fuses the well-optimized prior from these distinct domains cost-effectively and efficiently based on model interpolation. By integrating these two modules, PBaSR achieves commendable performance on both general and blur data without any additional inference and deployment cost and is generalizable across multiple model architectures. Rich experiments show that PBaSR achieves state-of-the-art (SOTA) performance across various quantitative metrics without incurring extra inference costs. Within the commonly adopted Learned …

Blind deblurring is an ill-posed inverse problem involving the retrieval of a clear image and blur kernel from a single blurry image. The challenge arises considerably when strong noise, where its level remains unknown, is introduced. Existing blind deblurring approaches heavily depend on designed priors for natural images and blur kernels. However, these methods are highly sensitive to noise due to the disturbance of the solution space. Here, we propose a noise-robust blind deblurring framework based on a novel kernel estimation function and deep image prior (DIP). Specifically, the proposed kernel estimation function mitigates noise and acquires the blur kernel, leveraging the capability of DIP to capture the priors of natural images. Additionally, the multiple kernel estimation scheme enables the successful execution of the deblurring task when the noise level is unknown. Extensive experimental studies, including simulated images and real-world examples, demonstrate the superior deblurring performance of the proposed method.

Recent years have seen significant advancements in image restoration, largely attributed to the development of modern deep neural networks, such as CNNs and Transformers. However, existing restoration backbones often face the dilemma between global receptive fields and efficient computation, hindering their application in practice. Recently, the Selective Structured State Space Model, especially the improved version Mamba, has shown great potential for long-range dependency modeling with linear complexity, which offers a way to resolve the above dilemma. However, the standard Mamba still faces certain challenges in low-level vision such as local pixel forgetting and channel redundancy. In this work, we introduce a simple but effective baseline, named MambaIR, which introduces both local enhancement and channel attention to improve the vanilla Mamba. In this way, our MambaIR takes advantage of the local pixel similarity and reduces the channel redundancy. Extensive experiments demonstrate the superiority of our method, for example, MambaIR outperforms SwinIR by up to 0.45dB on image SR, using similar computational cost but with a global receptive field.

Developing blind video deflickering (BVD) algorithms to enhance video temporal consistency, is gaining importance amid the flourish of image processing and video generation. However, the intricate nature of video data complicates the training of deep learning methods, leading to high resource consumption and instability, notably under severe lighting flicker. This underscores the critical need for a compact representation beyond pixel values to advance BVD research and applications. Inspired by the classic scale-time equalization (STE), our work introduces the histogram-assisted solution, called BlazeBVD, for high-fidelity and rapid BVD. Compared with STE, which directly corrects pixel values by temporally smoothing color histograms, BlazeBVD leverages smoothed illumination histograms within STE filtering to ease the challenge of learning temporal data using neural networks. In technique, BlazeBVD begins by condensing pixel values into illumination histograms that precisely capture flickering and local exposure variations. These histograms are then smoothed to produce singular frames set, filtered illumination maps, and exposure maps. Resorting to these deflickering priors, BlazeBVD utilizes a 2D network to restore faithful and consistent texture impacted by lighting changes or localized exposure issues. BlazeBVD also incorporates a lightweight 3D network to amend slight temporal inconsistencies, avoiding the resource consumption issue. Comprehensive experiments on synthetic, real-world …

Quantization is among the most common strategies to accelerate neural networks (NNs) on terminal devices. We are interested in increasing the robustness of Super Resolution (SR) networks to low-bit quantization considering mathematical model of natural images. Natural images contain partially smooth areas with edges between them. The number of pixels corresponding to edges is significantly smaller than the overall number of pixels. As SR task could be considered as ill-posed restoration of edges and texture, we propose to manually focus quantized CNNs on high-frequency part of the input image thus hiding quantization error in edges and texture providing visually appealing results. We extract edges and texture using well-known edge detectors based on finite-difference approximations of differential operators. To perform inverse transformation we propose to use solver for partial differential equations with regularization term that significantly increase solution robustness to errors in operator domain. The proposed approach significantly outperforms regular quantization counterpart in the case of full 4-bit quantization, for example, we achieved +3.76 dB for EDSR x2 and +3.67 dB for RFDN x2 on test part of DIV2K.

Optic deconvolution in light microscopy (LM) refers to recovering the object details from images, revealing the ground truth of samples. Traditional explicit methods in LM rely on the point spread function (PSF) during image acquisition. Yet, these approaches often fall short due to inaccurate PSF models and noise artifacts, hampering the overall restoration quality. In this paper, we approached the optic deconvolution as an inverse problem. Motivated by the nonstandard-form compression scheme introduced by Beylkin, Coifman, and Rokhlin (BCR), we proposed an innovative physics-informed neural network Multi-Stage Residual-BCR Net (m-rBCR) to approximate the optic deconvolution. We validated the m-rBCR model on four microscopy datasets - two simulated microscopy datasets from ImageNet and BioSR, real dSTORM microscopy images, and real widefield microscopy images. In contrast to the explicit deconvolution methods (e.g. Richardson-Lucy) and other state-of-the-art NN models (U-Net, DDPM, CARE, DnCNN, ESRGAN, RCAN, Noise2Noise, MPRNet, and MIMO-U-Net), the m-rBCR model demonstrates superior performance to other candidates by PSNR and SSIM in two real microscopy datasets and the simulated BioSR dataset. In the simulated ImageNet dataset, m-rBCR ranks the second-best place (right after MIMO-U-Net). With the backbone from the optical physics, m-rBCR exploits the trainable parameters with better performances (from ~30 times …

Hyperspectral image (HSI) reconstruction is vital for recovering spatial-spectral information from compressed measurements in coded aperture snapshot spectral imaging (CASSI) systems. Despite the effectiveness of end-to-end and deep unfolding methods, their reliance on substantial training data poses challenges, notably the scarcity of labeled HSIs. Existing approaches often train on limited datasets, such as KAIST and CAVE, leading to biased models with poor generalization capabilities. Addressing these challenges, we propose a universal Self-Supervised Adapter for Hyperspectral Snapshot Compressive Imaging (SAH-SCI). Unlike full fine-tuning or linear probing, SAH-SCI enhances model generalization by training a lightweight adapter while preserving the original model's parameters. We propose a novel approach that combines spectral and spatial adaptation to enhance an image model's capacity for spatial-spectral reasoning. Additionally, we introduce a customized adapter self-supervised loss function that captures the consistency, group invariance and image uncertainty of CASSI imaging. This approach effectively reduces the solution space for ill-posed HSI reconstruction. Experimental results demonstrate SAH's superiority over previous methods with fewer parameters, offering simplicity and adaptability to any end-to-end or unfolding methods. Our approach paves the way for leveraging more robust image foundation models in future hyperspectral imaging tasks.

Compressed Sensing (CS) facilitates rapid image acquisition by selecting a small subset of measurements sufficient for high-fidelity reconstruction. Adaptive CS seeks to further enhance this process by dynamically choosing future measurements based on information gleaned from data that is already acquired. However, many existing frameworks are often tailored to specific tasks and require intricate training procedures. We propose AdaSense, a novel Adaptive CS approach that leverages zero-shot posterior sampling with pre-trained diffusion models. By sequentially sampling from the posterior distribution, we can quantify the uncertainty of each possible future linear measurement throughout the acquisition process. AdaSense eliminates the need for additional training and boasts seamless adaptation to diverse domains with minimal tuning requirements. Our experiments demonstrate the effectiveness of AdaSense in reconstructing facial images from a small number of measurements. Furthermore, we apply AdaSense for active acquisition of medical images in the domains of magnetic resonance imaging (MRI) and computed tomography (CT), highlighting its potential for tangible real-world acceleration.

Computed tomography (CT) is widely utilized in clinical settings because it delivers detailed 3D images of the human body. However, performing CT scans is not always feasible due to radiation exposure and limitations in certain surgical environments. As an alternative, reconstructing CT images from ultra-sparse X-rays offers a valuable solution and has gained significant interest in scientific research and medical applications. However, it presents great challenges as it is inherently an ill-posed problem, often compromised by artifacts resulting from overlapping structures in X-ray images. In this paper, we propose DiffuX2CT, which models CT reconstruction from orthogonal biplanar X-rays as a conditional diffusion process. DiffuX2CT is established with a 3D global coherence denoising model with a new, implicit conditioning mechanism. We realize the conditioning mechanism by a newly designed tri-plane decoupling generator and an implicit neural decoder. By doing so, DiffuX2CT achieves structure-controllable reconstruction, which enables 3D structural information to be recovered from 2D X-rays, therefore producing faithful textures in CT images. As an extra contribution, we collect a real-world lumbar CT dataset, called LumbarV, as a new benchmark to verify the clinical significance and performance of CT reconstruction from X-rays. Extensive experiments on this dataset and three more publicly available …

Despite superior rate-distortion performance over traditional codecs, Neural Image Compression (NIC) is limited by its computational scalability in practical deployment. Prevailing research focuses on accelerating specific NIC modules but is restricted in controlling overall computational complexity. To this end, this work introduces BaSIC (BayesNet structure learning for computational Scalable neural Image Compression), a comprehensive, computationally scalable framework that affords full control over NIC processes. We learn the Bayesian network (BayesNet) structure of NIC for controlling both neural network backbones and autoregressive units. The learning of BayesNet is achieved by solving two sub-problems, i.e., learning a heterogeneous bipartite BayesNet for the inter-node structure to regulate backbone complexity, and a multipartite BayesNet for the intra-node structure to optimize parallel computation in autoregressive units. Experiments demonstrate that our method not only facilitates full computational scalability with more accurate complexity control but also maintains competitive compression performance compared to other computation scalable frameworks under equivalent computational constraints. Code will be available after acceptance.

Neural representation for video (NeRV), which employs a neural network to parameterize video signals, introduces a novel methodology in video representations. However, existing NeRV-based methods have difficulty in capturing fine spatial details and motion patterns due to spectral bias, in which a neural network learns high-frequency (HF) components at a slower rate than low-frequency (LF) components. In this paper, we propose spectra-preserving NeRV (SNeRV) as a novel approach to enhance implicit video representations by efficiently handling various frequency components. SNeRV uses 2D discrete wavelet transform (DWT) to decompose video into LF and HF features, preserving spatial structures and directly addressing the spectral bias issue. To balance the compactness, we encode only the LF components, while HF components that include fine textures are generated by a decoder. Specialized modules, including a multi-resolution fusion unit (MFU) and a high-frequency restorer (HFR), are integrated into a backbone to facilitate the representation. Furthermore, we extend SNeRV to effectively capture temporal correlations between adjacent video frames, by casting the extension as additional frequency decomposition to a temporal domain. This approach allows us to embed spatio-temporal LF features into the network, using temporally extended up-sampling blocks (TUBs). Experimental results demonstrate that SNeRV outperforms existing NeRV models …

Texture recognition has predominantly relied on methods based on handcrafted features and more recently, on Convolutional Neural Network (CNN)-based methods. However, many of these approaches do not capture the underlying directional relationships between visual vocabularies, attributes and features. In this study, we introduce a graph-based deep learning framework for texture and material recognition called Graph Texture Network (GTN) that models the underlying directional associations among latent texture attributes, that are hierarchically related to visual texture attributes, facilitating information exchange among them and consequently improving the discriminative capability among different texture and material categories. GTN, which is a non-Euclidean data structure, provides flexibility to learn complex underlying relationships among latent texture attributes via a learnable masked adjacency matrix. To ensure robustness of GTN to noise, especially on graphs with fewer vertices, we facilitate re-calibration of self-loop edge weights to preserve salient texture information within each vertex. We then utilize residual message passing to enrich the representations of latent texture attributes. Furthermore, GTN is able to facilitate interaction across multiple graphs, representing texture information across a range of scales. Finally, GTN can be easily incorporated into a variety of CNN architectures for end-to-end training and does not require fine-tuning of pre-trained CNN …

Offline signature verification (OSV) is a frequently utilized technology in forensics. This paper proposes a new model, DetailSemNet, for OSV. Unlike previous methods that rely on holistic features for pair comparisons, our approach underscores the significance of fine-grained differences for robust OSV. We propose to match local structures between two signature images, significantly boosting verification accuracy. Furthermore, we observe that without specific architectural modifications, transformer-based backbones might naturally obscure local details, adversely impacting OSV performance. To address this, we introduce a Detail-Semantics Integrator, leveraging feature disentanglement and re-entanglement. This integrator is specifically designed to enhance intricate details while simultaneously expanding discriminative semantics, thereby augmenting the efficacy of local structural matching. We evaluate our method against leading benchmarks in offline signature verification. Our model consistently outperforms recent methods, achieving state-of-the-art results with clear margins. The emphasis on local structure matching not only improves performance but also enhances the model's interpretability, supporting our findings. Additionally, our model demonstrates remarkable generalization capabilities in cross-dataset testing scenarios. The combination of generalizability and interpretability significantly bolsters the potential of DetailSemNet for real-world applications.

In recent years, the study of adversarial robustness in object detection systems, particularly those based on deep neural networks (DNNs), has become a pivotal area of research. Traditional physical attacks targeting object detectors, such as adversarial patches and texture manipulations, directly manipulate the surface of the object. While these methods are effective, their overt manipulation of objects may draw attention in real-world applications. To address this, this paper introduces a more subtle approach: an inconspicuous adversarial trigger that operates outside the bounding boxes, rendering the object undetectable to the model. We further enhance this approach by proposing the Feature Guidance (FG) technique and the Universal Auto-PGD (UAPGD) optimization strategy for crafting high-quality triggers. The effectiveness of our method is validated through extensive empirical testing, demonstrating its high performance in both digital and physical environments.

Previous studies in deepfake detection have shown promising results when testing face forgeries from the same dataset as the training. However, the problem remains challenging when one tries to generalize the detector to forgeries from unseen datasets and created by unseen methods. Utilizing forgery augmentation is one of the powerful lines to improve generalization performances. In this work, we present a novel general deepfake detection method, called Curricular Dynamic Forgery Augmentation (CDFA), which jointly trains a deepfake detector with a forgery augmentation policy network. Unlike the previous works, we propose to progressively apply forgery augmentations following a monotonic curriculum during the training. We further propose a dynamic forgery searching strategy to select one suitable forgery augmentation operation for each image varying between training stages, producing a forgery augmentation policy optimized for better generalization. In addition, we propose a novel forgery augmentation named self-shifted blending image to simply imitate the temporal inconsistency of deepfake generation. Comprehensive experiments show that CDFA can significantly improve both cross-datasets and cross-manipulations performances of various naive deepfake detectors in a plug-and-play way, and make them attain superior performances over the existing methods in several benchmark datasets. To facilitate the reproducible research, we will release our code …

Face recognition (FR) systems are vulnerable to external information leakage (EIL) attacks, which can reveal sensitive information about the training data, thus compromising the confidentiality of the company's proprietary and the privacy of the individuals concerned. Existing EIL attacks mainly rely on unrealistic assumptions, such as a high query budget for the attacker and massive computational power, resulting in impractical EIL attacks. We present AdversariaLeak, a novel and practical query-based EIL attack that targets the face verification model of the FR systems by using carefully selected adversarial samples. AdversariaLeak uses substitute models to craft adversarial samples, which are then handpicked to infer sensitive information. Our extensive evaluation on the MAAD-Face and CelebA datasets, which includes over 200 different target models, shows that AdversariaLeak outperforms state-of-the-art EIL attacks in inferring the property that best characterizes the FR model's training set while maintaining a small query budget and practical attacker assumptions.

Remote photoplethysmography (rPPG) has gained increasing attention in recent years for its ability to extract physiological signals from facial videos. Existing rPPG measurement methods have shown satisfactory performance in the intra-dataset and cross-dataset scenarios. However, they often neglect the incremental learning scenario, in which training data is presented sequentially, resulting in the issue of catastrophic forgetting. Meanwhile, mainstream class incremental learning algorithms suffer performance degradation or even fail to effectively transfer to rPPG measurement. In this paper, we present a novel and practical method to tackle continual learning for rPPG measurement. We firstly employ adapter finetuning to adapt to new tasks efficiently while enhancing the model's stability. To alleviate catastrophic forgetting without storing previous samples, we design a prototype-based augmentation strategy to reproduce the domain factors of previous tasks. Additionally, drawing inspiration from humans' problem-solving manner, an inference simplification strategy is devised to convert the potentially forgotten tasks into familiar ones for the model. To evaluate our method and enable fair comparisons, we create the first continual learning protocol for rPPG measurement. Extensive experiments demonstrate that our approach significantly surpasses the state-of-the-art methods.

Recent fMRI-to-image approaches mainly focused on associating fMRI signals with specific conditions of pre-trained diffusion models. These approaches, while producing high-quality images, capture only a limited aspect of the complex information in fMRI signals and offer little detailed control over image creation. In contrast, this paper proposes to directly modulate the generation process of diffusion models using fMRI signals. Our approach, NeuroPictor, divides the fMRI-to-image process into three steps: i) fMRI calibrated-encoding, to tackle multi-individual pre-training for a shared latent space to minimize individual difference and enable the subsequent cross-subject training; ii) fMRI-to-image cross-subject pre-training, perceptually learning to guide diffusion model with high- and low-level conditions across different individuals; iii) fMRI-to-image single-subject refining, similar with step ii but focus on adapting to particular individual. NeuroPictor extracts high-level semantic features from fMRI signals that characterizing the visual stimulus and incrementally fine-tunes the diffusion model with a low-level manipulation network to provide precise structural instructions. By training with over 60,000 fMRI-image pairs from various individuals, our model enjoys superior fMRI-to-image decoding capacity, particularly in the within-subject setting, as evidenced in benchmark datasets.

In this study, we address the intricate challenge of multi-task dense prediction, encompassing tasks such as semantic segmentation, depth estimation, and surface normal estimation, particularly when dealing with partially annotated data (MTPSL). The complexity arises from the absence of complete task labels for each training image. Given the inter-related nature of these pixel-wise dense tasks, our focus is on mining and capturing cross-task relationships. Existing solutions typically rely on learning global image representations for global cross-task image matching, imposing constraints that, unfortunately, sacrifice the finer structures within the images. Attempting local matching as a remedy faces hurdles due to the lack of precise region supervision, making local alignment a challenging endeavor. The introduction of Segment Anything Model (SAM) sheds light on addressing local alignment challenges by providing free and high-quality solutions for region detection. Leveraging SAM-detected regions, the subsequent challenge lies in aligning the representations within these regions. Diverging from conventional methods that directly learn a monolithic image representation, our proposal involves modeling region-wise representations using Gaussian Distributions. Aligning these distributions between corresponding regions from different tasks imparts higher flexibility and capacity to capture intra-region structures, accommodating a broader range of tasks. This innovative approach significantly enhances our ability to …

In this work, we describe a method for large-scale 3D cell-tracking through a segmentation selection approach. The proposed method is effective at tracking cells across large microscopy datasets on two fronts: (i) It can solve problems containing millions of segmentation instances in terabyte-scale 3D+t datasets; (ii) It achieves competitive results with or without deep learning, bypassing the requirement of 3D annotated data, that is scarce in the fluorescence microscopy field. The proposed method computes cell tracks and segments using a hierarchy of segmentation hypotheses and selects disjoint segments by maximizing the overlap between adjacent frames. We show that this method is the first to achieve state-of-the-art in both nuclei- and membrane-based cell tracking by evaluating it on the 2D epithelial cell benchmark and 3D images from the cell tracking challenge. Furthermore, it has a faster integer linear programming formulation, and the framework is flexible, supporting segmentations from individual off-the-shelf cell segmentation models or their combination as an ensemble. The code is available as supplementary material.

Knowing merely where the target is located is not sufficient for many real-life scenarios. In contrast, capturing rich details about the tracked target via its semantic trajectory, i.e. who/what this target is interacting with and when, where, and how they are interacting over time, is especially crucial and beneficial for various applications (e.g., customer analytics, public safety). We term such tracking as Semantic Tracking and define it as tracking the target based on the user's input and then, most importantly, capturing the semantic trajectory of this target. Acquiring such information can have significant impacts on sales, public safety, etc. However, currently there is no dataset for such comprehensive tracking of the target. To address this gap, we create SemTrack, a large and comprehensive dataset containing annotations of the target's semantic trajectory. The dataset contains 6.7 million frames from 6961 videos, covering a wide range of 52 different interaction classes with 115 different object classes spanning 10 different supercategories in 12 types of different scenes, including both indoor and outdoor environments. We also propose SemTracker, a simple and effective method, and incorporate a meta-learning approach to better handle the challenges of this task. Our dataset and code can be found at …

Neuromorphic sensors, specifically event cameras, revolutionize visual data acquisition by capturing pixel intensity changes with exceptional dynamic range, minimal latency, and energy efficiency, setting them apart from conventional frame-based cameras. The distinctive capabilities of event cameras have ignited significant interest in the domain of event-based action recognition, recognizing their vast potential for advancement. However, the development in this field is currently slowed by the lack of comprehensive, large-scale datasets, which are critical for developing robust recognition frameworks. To bridge this gap, we introduces DailyDVS-200, a meticulously curated benchmark dataset tailored for the event-based action recognition community. DailyDVS-200 is extensive, covering 200 action categories across real-world scenarios, recorded by 47 participants, and comprises more than 22,000 event sequences. This dataset is designed to reflect a broad spectrum of action types, scene complexities, and data acquisition diversity. Each sequence in the dataset is annotated with 14 attributes, ensuring a detailed characterization of the recorded actions. Moreover, DailyDVS-200 is structured to facilitate a wide range of research paths, offering a solid foundation for both validating existing approaches and inspiring novel methodologies. By setting a new benchmark in the field, we challenge the current limitations of neuromorphic data processing and invite a surge of …

Group activity detection (GAD) is the task of identifying members of each group and classifying the activity of the group at the same time in a video. While GAD has been studied recently, there is still much room for improvement in both dataset and methodology due to their limited capability to address practical GAD scenarios. To resolve these issues, we first present a new dataset, dubbed Café. Unlike existing datasets, Café is constructed primarily for GAD and presents more practical scenarios and metrics, as well as being large-scale and providing rich annotations. Along with the dataset, we propose a new GAD model that deals with an unknown number of groups and latent group members efficiently and effectively. We evaluated our model on three datasets including Café, where it outperformed previous work in terms of both accuracy and inference speed.

We consider the problem of transferring a temporal action segmentation system initially designed for exocentric (fixed) cameras to an egocentric scenario, where wearable cameras capture video data. The conventional supervised approach requires the collection and labeling of a new set of egocentric videos to adapt the model, which is costly and time-consuming. Instead, we propose a novel methodology which performs the adaptation leveraging existing labeled exocentric videos and a new set of unlabeled, synchronized exocentric-egocentric video pairs, for which temporal action segmentation annotations do not need to be collected. We implement the proposed methodology with an approach based on knowledge distillation, which we investigate both at the feature and model level. To evaluate our approach, we introduce a new benchmark based on the Assembly101 dataset. Results demonstrate the feasibility and effectiveness of the proposed method against classic unsupervised domain adaptation and temporal sequence alignment approaches. Remarkably, without bells and whistles, our best model performs on par with supervised approaches trained on labeled egocentric data, without ever seeing a single egocentric label, achieving a +15.99% (28.59% vs 12.60%) improvement in the edit score on the Assembly101 dataset compared to a baseline model trained solely on exocentric data.

Online temporal action localization (On-TAL) is the task of identifying multiple action instances given a streaming video. Since existing methods take as input only a video segment of fixed size per iteration, they are limited in considering long-term context and require tuning the segment size carefully. To overcome these limitations, we propose memory-augmented transformer (MATR). MATR utilizes the memory queue that selectively preserves the past segment features, allowing to leverage long-term context for inference. We also propose a novel action localization method that observes the current input segment to predict the end time of the current action and accesses the memory queue to estimate start time of the action. Our method outperformed existing methods on two datasets, THUMOS14 and MUSES, surpassing not only TAL methods in the online setting but also some offline TAL methods.

In this paper we present EgoLifter, a novel system that can automatically segment scenes captured from egocentric sensors into a complete decomposition of individual 3D objects. The system is specifically designed for egocentric data where scenes contain hundreds of objects captured from natural (non-scanning) motion. EgoLifter adopts 3D Gaussians as the underlying representation of 3D scenes and objects and uses segmentation masks from the Segment Anything Model (SAM) as weak supervision to learn flexible and promptable definitions of object instances free of any specific object taxonomy. To handle the challenge of dynamic objects in ego-centric videos, we design a transient prediction module that learns to filter out dynamic objects in the 3D reconstruction. The result is a fully automatic pipeline that is able to reconstruct 3D object instances as collections of 3D Gaussians that collectively compose the entire scene. We created a new benchmark on the Aria Digital Twin dataset that quantitatively demonstrates its state-of-the-art performance in open-world 3D segmentation from natural egocentric input. We run EgoLifter on various egocentric activity datasets which shows the promise of the method for 3D egocentric perception at scale.

We present MeshSegmenter, a pioneering framework designed for 3D Zero-shot semantic segmentation. This innovative model successfully extends the powerful capabilities of 2D segmentation models to 3D meshes, delivering accurate 3D segmentation across diverse meshes and segment descriptions. Our contributions are threefold. Firstly, we introduce the MeshSegmenter framework, which consistently produces precise 3D segmentation results. Secondly, we propose the generation of textures based on object descriptions to augment 2D segmentation models with additional texture information, thereby improving their accuracy. By harnessing latent texture information unearthed from generative models based on 3D meshes, our model can perform accurate 3D segmentation in geometrically non-prominent areas, such as segmenting a car door within a car mesh. Lastly, we develop a multi-view revoting module that integrates 2D detection results and confidence scores from various views onto the 3D mesh, ensuring the 3D consistency of segmentation results and eliminating inaccuracies from specific perspectives. Through these innovations, MeshSegmenter offers stable and reliable 3D segmentation results, highlighting its potential as a transformative tool in the field of 3D zero-shot segmentation.

Existing semi-supervised video object segmentation methods either focus on temporal feature matching or spatial-temporal feature modeling. However, they do not address the issues of sufficient target interaction and efficient parallel processing simultaneously, thereby constraining the learning of dynamic, target-aware features. To tackle these limitations, this paper proposes a spatial-temporal multi-level association framework, which jointly associates reference frame, test frame, and object features to achieve sufficient interaction and parallel target ID association with a spatial-temporal memory bank for efficient video object segmentation. Specifically, we construct a spatial-temporal multi-level feature association module to learn better target-aware features, which formulates feature extraction and interaction as the efficient operations of object self-attention, reference object enhancement, and test reference correlation. In addition, we propose a spatial-temporal memory to assist feature association and temporal ID assignment and correlation. We evaluate the proposed method by conducting extensive experiments on numerous video object segmentation datasets, including DAVIS 2016/2017 val, DAVIS 2017 test-dev, and YouTube-VOS 2018/2019 val. The favorable performance against the state-of-the-art methods demonstrates the effectiveness of our approach.

Long-term action anticipation has become an important task for many applications such as autonomous driving and human-robot interaction. Unlike short-term anticipation, predicting more actions into the future imposes a real challenge with the increasing uncertainty in longer horizons. While there has been a significant progress in predicting more actions into the future, most of the proposed methods address the task in a deterministic setup and ignore the underlying uncertainty. In this paper, we propose a novel temporal diffusion network that models the uncertainty of both the observation and the future predictions. As generator, we introduce a Gated Anticipation Network (GTAN) to model both observed and unobserved frames of a video in a mutual representation. On the one hand, using a mutual representation for past and future allows us to jointly model ambiguities in the observation and future, while on the other hand by design GTAN can intrinsically treat the observed and unobserved parts differently and steer the information flow between them. Our model achieves state-of-the-art results on the Breakfast, Assembly101 and 50Salads datasets in both stochastic and deterministic settings.

We propose ViC-MAE, a model that combines both Masked AutoEncoders (MAE) and contrastive learning. ViC-MAE is trained using a global feature obtained by pooling the local representations learned under an MAE reconstruction loss and leveraging this representation under a contrastive objective across images and video frames. We show that visual representations learned under ViC-MAE generalize well to video and image classification tasks. Particularly, ViC-MAE obtains state-of-the-art transfer learning performance from video to images on Imagenet-1k compared to the recently proposed OmniMAE by achieving a top-1 accuracy of 86% (+1.3% absolute improvement) when trained on the same data and 87.1% (+2.4% absolute improvement) when training on extra data. At the same time, ViC-MAE outperforms most other methods on video benchmarks by obtaining 75.9% top-1 accuracy on the challenging Something something-v2 video benchmark. When training on videos and images from diverse datasets, our method maintains a balanced transfer-learning performance between video and image classification benchmarks, coming only as a close second to the best-supervised method. Source code and model checkpoints will be released with this paper.

Action quality assessment (AQA) is a challenging vision task that requires discerning and quantifying subtle differences in actions from the same class. While recent research has made strides in creating fine-grained annotations for more precise analysis, existing methods primarily focus on coarse action segmentation, leading to limited identification of discriminative action frames. To address this issue, we propose a Vision-Language Action Knowledge Learning approach for action quality assessment, along with a multi-grained alignment framework to understand different levels of action knowledge. In our framework, prior knowledge, such as specialized terminology, is embedded into video-level, stage-level, and frame-level representations via CLIP. We further propose a new semantic-aware collaborative attention module to prevent confusing interactions and preserve textual knowledge in cross-modal and cross-semantic spaces. Specifically, we leverage the powerful cross-modal knowledge of CLIP to embed textual semantics into image features, which then guide action spatial-temporal representations. Our approach can be plug-and-played with existing AQA methods, frame-wise annotations or not. Extensive experiments and ablation studies show that our approach achieves state-of-the-art on four public short and long-term AQA benchmarks: FineDiving, MTL-AQA, JIGSAWS, and Fis-V.

Addressing the dual challenges of local redundancy and global dependencies in video understanding, this work innovatively adapts the Mamba to the video domain. The proposed VideoMamba overcomes the limitations of existing 3D convolution neural networks and video transformers. Its linear-complexity operator enables efficient long-term modeling, which is crucial for high-resolution long video understanding. Extensive evaluations reveal VideoMamba's four core abilities: (1) Scalability in the visual domain without extensive dataset pretraining, thanks to a novel self-distillation technique; (2) Sensitivity for recognizing short-term actions even with fine-grained motion differences; (3) Superiority in long-term video understanding, showcasing significant advancements over traditional feature-based models; and (4) Compatibility with other modalities, demonstrating robustness in multi-modal contexts. Through these distinct advantages, VideoMamba sets a new benchmark for video understanding, offering a scalable and efficient solution for comprehensive video understanding. All the code and models will be released.

Videos are highly redundant data source and it is often enough to identify a few key moments to solve any given task. In this paper, we present a text-conditioned video resampler (TCR) module that uses a pre-trained and frozen visual encoder and large language model (LLM) to process long video sequences for a task. TCR localises relevant visual features from the video given a text condition and provides them to a LLM to generate a text response. Due to its lightweight design and use of cross-attention, TCR can process more than 100 frames at a time allowing the model to use much longer chunks of video than earlier works. We make the following contributions: (i) we design a transformer-based sampling architecture that can process long videos conditioned on a task, together with a training method that enables it to bridge pre-trained visual and language models; (ii) we empirically validate its efficacy on a wide variety of evaluation tasks, and set a new state-of-the-art on NextQA, EgoSchema, and the EGO4D-LTA challenge; and (iii) we determine tasks which require longer video contexts and that can thus be used effectively for further evaluation of long-range video models.

Temporal grounding, also known as video moment retrieval, aims at locating video segments corresponding to a given query sentence. The compositional nature of natural language enables the localization beyond predefined events, posing a certain challenge to the compositional generalizability of existing methods. Recent studies establish the correspondence between videos and queries through a decompose-reconstruct manner, utilizing contrastive learning to achieve compositional generalization. However, these approaches have significant drawbacks in constructing negative queries: (1) Only dominant verbs and nouns are considered, overlooking the impact of other primitives such as prepositions and adverbs. (2) These negative samples are formed through random sampling and recombination, resulting in numerous semantically implausible combinations and forcing the model to learn unrealistic semantic differences. To address these limitations, we first propose a large language model-driven method for negative query construction, utilizing GPT-3.5-Turbo to generate semantically plausible hard negative queries. Subsequently, we introduce a coarse-to-fine saliency ranking strategy, which encourages the model to learn the multi-granularity semantic relationships between videos and hierarchical negative queries to boost compositional generalization. Extensive experiments on two challenging benchmarks, Charades-CG and ActivityNet-CG, validate the effectiveness and generalizability of our proposed method. Our code will be available soon.

What makes good representations for video understanding, such as anticipating future activities, or answering video-conditioned questions? While earlier approaches focus on end-to-end learning directly from video pixels, we propose to revisit text-based representations, such as general-purpose video captions, which are interpretable and can be directly consumed by large language models (LLMs). Intuitively, different video understanding tasks may require representations that are complementary and at different granularity. To this end, we propose versatile action models (Vamos), a learning framework powered by a large language model as the ``reasoner'', and can flexibly leverage visual embedding and free-form text descriptions as its input. To interpret the important text evidence for question answering, we generalize the concept bottleneck model to work with tokens and nonlinear models, which uses hard attention to select a small subset of tokens from the free-form text as inputs to the LLM reasoner. We evaluate Vamos on four complementary video understanding benchmarks, Ego4D, Next-QA, IntentQA, and EgoSchema, on its capability to model temporal dynamics, encode visual history, and perform reasoning. Surprisingly, we observe that text-based representations consistently achieve competitive performance on all benchmarks, and that visual embeddings provide marginal or no performance improvement, demonstrating the effectiveness of text-based video representation …

Most current LLM-based models for video understanding can process videos within minutes. However, they struggle with lengthy videos due to challenges such as “noise and redundancy", as well as “memory and computation" constraints. In this paper, we present Goldfish, a methodology tailored for comprehending videos of arbitrary lengths. We also introduce the TVQA-long benchmark, specifically designed to evaluate models’ capabilities in understanding long videos with questions in both vision and text content. Goldfish approaches these challenges with an efficient retrieval mechanism that initially gathers the top-k video clips relevant to the instruction before proceeding to provide the desired response. This design of the retrieval mechanism enables the Goldfish to efficiently process arbitrarily long video sequences, facilitating its application in contexts such as movies or television series. To facilitate the retrieval process, we developed MiniGPT4-Video that generates detailed descriptions for the video clips. In addressing the scarcity of benchmarks for long video evaluation, we adapted the TVQA short video benchmark for extended content analysis by aggregating questions from entire episodes, thereby shifting the evaluation from partial to full episode comprehension. We attained a 41.78% accuracy rate on the TVQA-long benchmark, surpassing previous methods by 14.94%. Our MiniGPT4-Video also shows exceptional performance …

Data quality stands at the forefront of deciding the effectiveness of video-language representation learning. However, video-text pairs in previous data typically do not align perfectly with each other, which might lead to video-language representations that do not accurately reflect cross-modal semantics. Moreover, previous data also possess an uneven distribution of concepts, thereby hampering the downstream performance across unpopular subjects. To address these problems, we propose a contrastive objective with a subtractive angular margin to regularize cross-modal representations in their effort to reach perfect similarity. Furthermore, to adapt to the non-uniform concept distribution, we propose a multi-layer perceptron (MLP)-parameterized weighting function that maps loss values to sample weights which enable dynamic adjustment of the model’s focus throughout the training. With the training guided by a small amount of unbiased meta-data and augmented by video-text data generated by large vision-language model, we improve video-language representations and achieve superior performances on commonly used video question answering and text-video retrieval datasets.

In this paper, we aim to establish an automatic, scalable pipeline for denoising the large-scale instructional dataset and construct a high-quality video-text dataset with multiple descriptive steps supervision, named HowToStep. We make the following contributions: (i) improving the quality of sentences in dataset by upgrading ASR systems to reduce errors from speech recognition and prompting a large language model to transform noisy ASR transcripts into descriptive steps; (ii) proposing a Transformer-based architecture with all texts as queries, iteratively attending to the visual features, to temporally align the generated steps to corresponding video segments. To measure the quality of our curated datasets, we train models for the task of multi-sentence grounding on it, i.e., given a long-term video, and associated multiple sentences, our goal is to determine their corresponding timestamps in the video simultaneously, as a result, the model demonstrates superior performance on a series of multi-sentence grounding tasks, surpassing existing state-of-the-art methods by a significant margin on three public benchmarks, namely, 9.0% on HT-Step, 5.1% on HTM-Align and 1.9% on CrossTask. All codes, models, and the resulting dataset will be publicly released to the research community.

This paper focuses on the challenge of answering questions in scenarios that are composed of rich and complex dynamic audio-visual components. Although existing Multimodal Large Language Models (MLLMs) can respond to audio-visual content, these responses are sometimes ambiguous and fail to describe specific audio-visual events. To overcome this limitation, we introduce the CAT, which enhances MLLM in three ways: 1) besides straightforwardly bridging audio and video, we design a clue aggregator that aggregates question-related clues in dynamic audio-visual scenarios to enrich the detailed knowledge required for large language models. 2) CAT is trained on a mixed multimodal dataset, allowing direct application in audio-visual scenarios. Notably, we collect an audio-visual joint instruction dataset named AVinstruct, to further enhance the capacity of CAT to model cross-semantic correlations. 3) we propose AI-assisted ambiguity-aware direct preference optimization, a strategy specialized in retraining the model to favor the non-ambiguity response and improve the ability to localize specific audio-visual objects. Extensive experimental results demonstrate that CAT outperforms existing methods on multimodal tasks, especially in Audio-Visual Question Answering (AVQA) tasks. The codes and the collected instructions will be released soon.

Audio-visual segmentation (AVS) is an emerging task that aims to accurately segment sounding objects based on audio-visual cues. The success of AVS learning systems depends on the effectiveness of cross-modal interaction. Such a requirement can be naturally fulfilled by leveraging transformer-based segmentation architecture due to its inherent ability to capture long-range dependencies and flexibility in handling different modalities. However, the inherent training issues of transformer-based methods, such as the low efficacy of cross-attention and unstable bipartite matching, can be amplified in AVS, particularly when the learned audio query does not provide a clear semantic clue. In this paper, we address these two issues with the new Class-conditional Prompting Machine (CPM). CPM improves the bipartite matching with a learning strategy combining class-agnostic queries with class-conditional queries. The efficacy of cross-modal attention is upgraded with new learning objectives for the audio, visual and joint modalities. We conduct experiments on AVS benchmarks, demonstrating that our method achieves state-of-the-art (SOTA) segmentation accuracy. Code will be available.

We present SignAvatars, the first large-scale, multi-prompt 3D sign language (SL) motion dataset designed to bridge the communication gap for Deaf and hard-of-hearing individuals. While there has been an exponentially growing number of research regarding digital communication, the majority of existing communication technologies primarily cater to spoken or written languages, instead of SL, the essential communication method for Deaf and hard-of-hearing communities. Existing SL datasets, dictionaries, and sign language production (SLP) methods are typically limited to 2D as annotating 3D models and avatars for SL is usually an entirely manual and labor-intensive process conducted by SL experts, often resulting in unnatural avatars. In response to these challenges, we compile and curate the SignAvatars dataset, which comprises 70,000 videos from 153 signers, totaling 8.34 million frames, covering both isolated signs and continuous, co-articulated signs, with multiple prompts including HamNoSys, spoken language, and words. To yield 3D holistic annotations, including meshes and biomechanically-valid poses of body, hands, and face, as well as 2D and 3D keypoints, we introduce an automated annotation pipeline operating on our large corpus of SL videos. SignAvatars facilitates various tasks such as 3D sign language recognition (SLR) and the novel 3D SL production (SLP) from diverse inputs like …

Video geolocalization is a crucial problem in current times. Given just a video, ascertaining where it is coming from can have a plethora of advantages. The problem of worldwide geolocalization has been tackled before, but only using the image modality. Its video counterpart remains relatively unexplored. Meanwhile, video geolocalization has also garnered some attention in the recent past, but the existing methods are all restricted to specific regions. This motivates us to explore the problem of video geolocalization at a global scale. Hence, we propose a novel problem of worldwide video geolocalization with the objective of hierarchically predicting the correct city, state/province, country, and continent, given a video. However, no large scale video datasets that have extensive worldwide coverage exist, to train models for solving this problem. To this end, we introduce a new dataset, "CityGuessr68k'' comprising of 68,269 videos from 166 cities all over the world. We also propose a novel baseline approach to this problem, by designing a transformer-based architecture comprising of an elegant "Self-Cross Attention'' module for incorporating scenes as well as a "TextLabel Alignment'' strategy for distilling knowledge from textlabels in feature space. To further enhance our location prediction, we also utilize soft-scene labels. Finally we …

For the visible-infrared person re-identification (VI-ReID) task, one of the primary challenges lies in significant cross-modality discrepancy. Existing methods struggle to conduct modality-invariant information mining. They often focus solely on mining singular dimensions like spatial or channel, and overlook the extraction of specific-modality multi-dimension information. To fully mine modality-invariant information across a wide range, we introduce the Wide-Ranging Information Mining Network (WRIM-Net), which mainly comprises a Multi-dimension Interactive Information Mining (MIIM) module and an Auxiliary-Information-based Contrastive Learning (AICL) approach. Empowered by the proposed Global Region Interaction (GRI), MIIM comprehensively mines non-local spatial and channel information through intra-dimension interaction. Moreover, Thanks to the low computational complexity design, separate MIIM can be positioned in shallow layers, enabling the network to better mine specific-modality multi-dimension information. AICL, by introducing the novel Cross-Modality Key-Instance Contrastive (CMKIC) loss, effectively guides the network in extracting modality-invariant information. We conduct extensive experiments not only on the well-known SYSU-MM01 and RegDB datasets but also on the latest large-scale cross-modality LLCM dataset. The results demonstrate WRIM-Net's superiority over state-of-the-art methods.

Multimodal image-text contrastive learning has shown that joint representations can be learned across modalities. Here, we show how leveraging multiple views of image data with contrastive learning can improve downstream fine-grained classification performance for species recognition, even when one view is absent. We propose ContRastive Image-remote Sensing Pre-training (CRISP)—a new pre-training task for ground-level and aerial image representation learning of the natural world—and introduce Nature Multi-View (NMV), a dataset of natural world imagery including >3 million ground-level and aerial image pairs for over 6,000 plant taxa across the ecologically diverse state of California. The NMV dataset and accompanying material are available at hf.co/datasets/andyvhuynh/NatureMultiView.

In this study, we introduce a new problem raised by social media and photojournalism, named Image Address Localization (IAL), which aims to predict the readable textual address where an image was taken. Existing two-stage approaches involve predicting geographical coordinates and converting them into human-readable addresses, which can lead to ambiguity and be resource-intensive. In contrast, we propose an end-to-end framework named AddressCLIP to solve the problem with more semantics, consisting of two key ingredients: i) image-text alignment to align images with addresses and scene captions by contrastive learning, and ii) image-geography matching to constrain image features with the spatial distance in terms of manifold learning. Additionally, we have built three datasets from Pittsburgh and San Francisco on different scales specifically for the IAL problem. Experiments demonstrate that our approach achieves compelling performance on the proposed datasets and outperforms representative transfer learning methods for vision-language models. Furthermore, extensive ablations and visualizations exhibit the effectiveness of the proposed method. The datasets and source code are available at https://github.com/xsx1001/AddressCLIP.

We introduce LingoQA, a novel dataset and benchmark for video question answering in autonomous driving. The dataset contains 28K unique short video scenarios, and 419K annotations. Evaluating state-of-the-art vision-language models on our benchmark shows that their performance is below human capabilities, with GPT-4V responding truthfully to 56.67% of the questions compared to 93.4% for humans. For evaluation, in addition to conducting a human study, we propose a truthfulness classifier, called Lingo-Judge, that achieves a 0.95 Spearman correlation coefficient to human evaluations, surpassing existing techniques like METEOR, BLEU, CIDEr, and GPT-4. We establish a baseline vision-language model and run extensive ablation studies to understand its performance. We release our dataset and benchmark as an evaluation platform for vision-language models in autonomous driving.

The quest for fully autonomous vehicles (AVs) capable of navigating complex real-world scenarios with human-like understanding and responsiveness. In this paper, we introduce Dolphins, a novel vision-language model architected to imbibe human-like abilities as a conversational driving assistant. Dolphins is adept at processing multimodal inputs comprising video (or image) data, text instructions, and historical control signals to generate informed outputs corresponding to the provided instructions. Building upon the open-sourced pretrained Vision-Language Model, OpenFlamingo, we first enhance Dolphins's reasoning capabilities through an innovative Grounded Chain of Thought (GCoT) process in the general domain. Then we tailored Dolphins to the driving domain by constructing driving-specific instruction data and conducting instruction tuning. Through the utilization of the BDD-X dataset, we designed and consolidated four distinct AV tasks into Dolphins to foster a holistic understanding of intricate driving scenarios. As a result, the distinctive features of Dolphins are characterized into two dimensions: (1) the ability to provide a comprehensive understanding of complex and long-tailed open-world driving scenarios and solve a spectrum of AV tasks, and (2) the emergence of human-like capabilities including gradient-free instant adaptation via in-context learning and error recovery via reflection.

Vision and language navigation is a task that requires an agent to navigate according to a natural language instruction. Recent methods predict sub-goals on constructed topology map at each step to enable long-term action planning. However, they suffer from high computational cost when attempting to support such high-level predictions with GCN-like models. In this work, we propose an alternative method that facilitates navigation planning by considering the alignment between instructions and directed fidelity trajectories, which refers to a path from the initial node to the candidate locations on a directed graph without detours. This planning strategy leads to an efficient model while achieving strong performance. Specifically, we introduce a directed graph to illustrate the explored area of the environment, emphasizing directionality. Then, we firstly define the trajectory representation as a sequence of directed edge features, which are extracted from the panorama based on the corresponding orientation. Ultimately, we assess and compare the alignment between instruction and different trajectories during navigation to determine the next navigation target. Our method outperforms previous SOTA method BEVBert on RxR dataset and is comparable on R2R dataset with only 9% computational cost.

Vision-and-Language Navigation (VLN) involves guiding an agent through indoor environments using human-provided textual instructions. Coarse-grained VLN, with short and high-level instructions, has gained popularity as it closely mirrors real-world scenarios. However, a significant challenge is these instructions are often too concise for agents to comprehend and act upon. Previous studies have explored allowing agents to seek assistance during navigation, but typically offer rigid support from pre-existing datasets or simulators. The advent of Large Language Models (LLMs) presents a novel avenue for aiding VLN agents. This paper introduces VLN-Copilot, a framework enabling agents to actively seek assistance when encountering confusion, with the LLM serving as a copilot to facilitate navigation. Our approach includes the introduction of a confusion score, quantifying the level of uncertainty in an agent's action decisions, while the LLM offers real-time detailed guidance for navigation. Experimental results on two coarse-grained VLN datasets demonstrate the efficacy of our method.

Generalization is a pivotal challenge for agents following natural language instructions. To approach this goal, we leverage a vision-language model (VLM) for visual grounding and transfer its vision-language knowledge into reinforcement learning (RL) for object-centric tasks, which makes the agent capable of zero-shot generalization to unseen objects and instructions. By visual grounding, we obtain an object-grounded confidence map for the target object indicated in the instruction. Based on this map, we introduce two routes to transfer VLM knowledge into RL. Firstly, we propose an object-grounded intrinsic reward function derived from the confidence map to more effectively guide the agent towards the target object. Secondly, the confidence map offers a more unified, accessible task representation for the agent's policy, compared to language embeddings. This enables the agent to process unseen objects and instructions through comprehensible visual confidence maps, facilitating zero-shot object-level generalization. Single-task experiments prove that our intrinsic reward significantly improves performance on challenging skill learning. In multi-task experiments, through testing on tasks beyond the training set, we show that the agent, when provided with the confidence map as the task representation, possesses better generalization capabilities than language-based conditioning.

Real-world multi-modal problems are rarely solved by a single machine learning model, and often require multi-step computational plans that involve stitching several models. Tool-augmented LLMs hold tremendous promise for automating the generation of such computational plans. However, the lack of standardized benchmarks for evaluating LLMs as planners for multi-step multi-modal tasks has prevented a systematic study of planner design decisions. Should LLMs generate a full plan in a single shot or step-by-step? Should they invoke tools directly with Python code or through structured data formats like JSON? Does feedback improve planning? To answer these questions and more, we introduce m&m’s: a benchmark containing 4K+ multi-step multi-modal tasks involving 33 tools that include multi-modal models, (free) public APIs, and image processing modules. For each of these task queries, we provide automatically generated plans using this realistic toolset. We further provide a high-quality subset of 1565 task plans that are human-verified and correctly executable. With m&m’s, we evaluate 6 popular LLMs with 2 planning strategies (multi-step vs. step-by-step planning), 2 plan formats (JSON vs. code), and 3 types of feedback (parsing/verification/execution). Finally, we summarize takeaways from our extensive experiments and provide practical recommendations for designing planners for m&m’s tasks.

Visual Programming (VP) has emerged as a powerful framework for Visual Question Answering (VQA). By generating and executing bespoke code for each question, these methods show advancement in leveraging Large Language Models (LLMs) for complex problem-solving. Despite their potential, existing VP methods generate all code in a single function, which does not fully utilize LLM's reasoning capacity and the modular adaptability of code. This results in code that is suboptimal in terms of both accuracy and interpretability. Inspired by human coding practices, we propose Recursive Visual Programming (RVP), which better harnesses the reasoning capacity of LLMs, provides modular code structure between code pieces, and assigns different return types for the sub-problems elegantly. RVP approaches VQA tasks with an top-down recursive code generation approach, allowing decomposition of complicated problems into smaller parts. We show RVP's efficacy through extensive experiments on benchmarks including VSR, COVR, GQA, and NextQA, underscoring the value of adopting human-like recursive and modular programming techniques for solving VQA tasks.

Large foundation models have recently emerged as a prominent focus of interest, attaining superior performance in widespread scenarios. Due to the scarcity of 3D data, many efforts have been made to adapt pre-trained transformers from vision to 3D domains. However, such 2D-to-3D approaches are still limited, due to the potential loss of spatial geometries and high computation cost. More importantly, their frameworks are mainly designed for 2D models, lacking a general any-to-3D paradigm. In this paper, we introduce Any2Point, a parameter-efficient method to empower any-modality large models (vision, language, audio) for 3D understanding. Given a frozen transformer from any source modality, we propose a 3D-to-any (1D or 2D) virtual projection strategy that correlates the input 3D points to the original 1D or 2D positions within the source modality. This mechanism enables us to assign each 3D token with a positional encoding paired with the pre-trained model, which avoids 3D geometry loss caused by the true projection and better motivates the transformer for 3D learning with 1D/2D positional priors. Then, within each transformer block, we insert an any-to-3D guided adapter module for parameter-efficient fine-tuning. The adapter incorporates prior spatial knowledge from the source modality to guide the local feature aggregation of …

We introduce a Depicted image Quality Assessment method (DepictQA), overcoming the constraints of traditional score-based methods. DepictQA allows for detailed, language-based, human-like evaluation of image quality by leveraging Multi-modal Large Language Models (MLLMs). Unlike conventional Image Quality Assessment (IQA) methods relying on scores, DepictQA interprets image content and distortions descriptively and comparatively, aligning closely with humans' reasoning process. To build the DepictQA model, we establish a hierarchical task framework, and collect a multi-modal IQA training dataset. To tackle the challenges of limited training data and multi-image processing, we propose to use multi-source training data and specialized image tags. These designs result in a better performance of DepictQA than score-based approaches on multiple benchmarks. Moreover, compared with general MLLMs, DepictQA can generate more accurate reasoning descriptive languages. Our work shows the research potential of multi-modal IQA tasks. Codes and datasets are available in https://depictqa.github.io.

Hallucination has been a major problem for large language models and remains a critical challenge when it comes to multimodality in which vision-language models (VLMs) have to deal with not just textual but also visual inputs. Despite rapid progress in VLMs, resources for evaluating and addressing multimodal hallucination are limited and mostly focused on evaluation. This work introduces HaloQuest, a novel visual question answering dataset that captures various aspects of multimodal hallucination such as false premises, insufficient contexts, and visual challenges. A novel idea from HaloQuest is to leverage synthetic images, apart from real ones, to enable dataset creation at scale. With over 7.7K examples spanning across a wide variety of categories, HaloQuest was designed to be both a challenging benchmark for VLMs and a fine-tuning dataset for advancing multimodal reasoning. Our experiments reveal that current models struggle with HaloQuest, with all open-source VLMs achieving below 36\% accuracy. On the other hand, fine-tuning on HaloQuest significantly reduces hallucination rates while preserving performance on standard reasoning tasks. Our results discover that benchmarking with generated images is highly correlated (r=0.97) with real images. Last but not least, we propose a novel Auto-Eval mechanism that is highly correlated with human raters (r=0.99) for …

The rapid progression of Text-to-Image (T2I) and Multimodal Large Language Models (MLLMs) has resulted in their widespread adoption across multiple computer vision and natural language processing tasks. However, a common mode of failure that persists across both classes of models is their inability to correctly reason over spatial relationships. To tackle this shortcoming, we develop the REVISION framework which improves and evaluates spatial fidelity in vision-language models. REVISION is a 3D rendering based pipeline that generates spatially accurate synthetic images, given a textual prompt. REVISION is an extendable framework, which currently supports 101 3D assets, 11 spatial relationships, all with diverse camera perspectives and backgrounds. Leveraging images from REVISION as additional guidance in a training-free manner consistently improves the spatial consistency of T2I models across all spatial relationships, achieving competitive performance on the VISOR and T2I-CompBench benchmarks. We also introduce the REVISION benchmark to evaluate the spatial reasoning abilities of MLLMs, and find that state-of-the-art models are not robust to complex spatial reasoning under adversarial settings. Our results and findings indicate that utilizing rendering-based frameworks is an efficient approach for developing reasoning-aware generative models.

The proliferation of machine learning models in critical decision making processes has underscored the need for bias discovery and mitigation strategies. Identifying the reasons behind a biased system is not straightforward, since in many occasions they are associated with hidden spurious correlations which are not easy to spot. Standard approaches rely on bias audits performed by analyzing model performance in pre-defined subgroups of data samples, usually characterized by common attributes like gender or ethnicity when it comes to people, or other specific attributes defining semantically coherent groups of images. However, it is not always possible to know a-priori the specific attributes defining the failure modes of visual recognition systems. Recent approaches propose to discover these groups by leveraging large vision language models, which enable the extraction of cross-modal embeddings and the generation of textual descriptions to characterize the subgroups where a certain model is underperforming. In this work, we argue that incorporating visual explanations (e.g. heatmaps generated via GradCAM or other approaches) can boost the performance of such bias discovery and mitigation frameworks. To this end, we introduce Visually Grounded Bias Discovery and Mitigation (ViG-Bias), a simple yet effective technique which can be integrated to a variety of existing frameworks …

Multimodal Large Language Models (MLLMs) demonstrate exceptional problem-solving capabilities, but few research studies aim to gauge the ability to generate visual instruction tuning data. This paper proposes to explore the potential of empowering MLLMs to generate data independently without relying on GPT-4. We introduce Genixer, a comprehensive data generation pipeline consisting of four key steps: (i) instruction data collection, (ii) instruction template design, (iii) empowering MLLMs, and (iv) data generation and filtering. Additionally, we outline two modes of data generation: task-agnostic and task-specific, enabling controllable output. We demonstrate that a synthetic VQA-like dataset trained with LLaVA1.5 enhances performance on 10 out of 12 multimodal benchmarks. Additionally, the grounding MLLM Shikra, when trained with a REC-like synthetic dataset, shows improvements on 7 out of 8 REC datasets. Through experiments and synthetic data analysis, our findings are: (1) current MLLMs can serve as robust data generators without assistance from GPT-4V; (2) MLLMs trained with task-specific datasets can surpass GPT-4V in generating complex instruction tuning data; (3) synthetic datasets enhance performance across various multimodal benchmarks and help mitigate model hallucinations.

With the rapid advancement of multimodal learning, pre-trained Vision-Language Models (VLMs) such as CLIP have demonstrated remarkable capacities in bridging the gap between visual and language modalities. However, these models remain vulnerable to adversarial attacks, particularly in the image modality, presenting considerable security risks. This paper introduces Adversarial Prompt Tuning (AdvPT), a novel technique to enhance the adversarial robustness of image encoders in VLMs. AdvPT innovatively leverages learnable text prompts and aligns them with adversarial image embeddings, to address the vulnerabilities inherent in VLMs without the need for extensive parameter training or modification of the model architecture. We demonstrate that AdvPT improves resistance against white-box and black-box adversarial attacks and exhibits a synergistic effect when combined with existing image-processing-based defense techniques, further boosting defensive capabilities. Comprehensive experimental analyses provide insights into adversarial prompt tuning, a novel paradigm devoted to improving resistance to adversarial images through textual input modifications, paving the way for future robust multimodal learning research. These findings open up new possibilities for enhancing the security of VLMs. Our code will be available upon publication of the paper.

In this work we discuss building performant Multimodal Large Language Models (MLLMs). In particular, we study the importance of various architecture components and data choices. Through careful and comprehensive ablations of the image encoder, the vision-language connector, and various pre-training data choices, we identify performant components as well as design lessons. For example, we demonstrate that for large-scale multimodal pre-training using a careful mix of image-caption, interleaved image-text, and text-only data is crucial for achieving state-of-the-art (SOTA) few-shot results across multiple benchmarks, compared to other published pre-training results. Further, we show that the image encoder together with image resolution and the image token count has substantial impact, while the vision-language connector design is of comparatively negligible importance. By scaling up the presented recipe, we build MM1, a family of multimodal models up to 30B parameters, that are SOTA in pre-training metrics and achieve competitive performance after supervised fine-tuning on a range of established multimodal benchmarks. Thanks to large-scale pre-training, MM1 enjoys appealing properties such as enhanced in-context learning, and multi-image reasoning, enabling few-shot chain-of-thought prompting.

Multi-label Intention Understanding (MIU) for images is a critical yet challenging domain, primarily due to the ambiguity of intentions leading to a resource-intensive annotation process. Current leading approaches are held back by the limited amount of labeled data. To mitigate the scarcity of annotated data, we leverage the Contrastive Language-Image Pre-training (CLIP) model, renowned for its proficiency in aligning textual and visual modalities. We introduce a novel framework, Intention Understanding with CLIP (IntCLIP), which utilizes a dual-branch approach. This framework exploits the 'Sight'-oriented knowledge inherent in CLIP to augment 'Semantic'-centric MIU tasks. Additionally, we propose Hierarchical Class Integration to effectively manage the complex layered label structure, aligning it with CLIP's nuanced sentence feature extraction capabilities. Our Sight-assisted Aggregation further refines this model by infusing the semantic feature map with essential visual cues, thereby enhancing the intention understanding ability. Through extensive experiments conducted on the standard MIU benchmark and other subjective tasks such as Image Emotion Recognition, IntCLIP clearly demonstrates superiority over current state-of-the-art techniques. The code and models will be released.

Current high-resolution vision-language models encode images as high-resolution image tokens and exhaustively take all these tokens to compute attention, which significantly increases the computational cost. To address this problem, we propose FlexAttention, a flexible attention mechanism for efficient high-resolution vision-language models. Specifically, a high-resolution image is encoded both as high-resolution tokens and low-resolution tokens, where only the low-resolution tokens and a few selected high-resolution tokens are utilized to calculate the attention map, which greatly shrinks the computational cost. The high-resolution tokens are selected via a high-resolution selection module which could retrieve tokens of relevant regions based on an input attention map. The selected high-resolution tokens are then concatenated to the low-resolution tokens and text tokens, and input to a hierarchical self-attention layer which produces an attention map that could be used for the next-step high-resolution token selection. The hierarchical self-attention process and high-resolution token selection process are performed iteratively for each attention layer. Experiments on multimodal benchmarks prove that our FlexAttention outperforms existing high-resolution VLMs (e.g., relatively ~9% in V* Bench, ~7% in TextVQA), while also significantly reducing the computational cost by nearly 40%.

Large language models are built on top of a transformer-based architecture to process textual inputs. For example, the LLaMA family of models stand out among many open-source implementations. Can the same transformer be used to process 2D images? In this paper, we answer this question by unveiling a LLaMA-like vision transformer in plain and pyramid forms, termed VisionLLaMA, which is tailored for this purpose. We extensively evaluate its effectiveness using typical pre-training paradigms in a good portion of downstream tasks of image perception and especially image generation. In many cases, VisionLLaMA has exhibited substantial gains over the previous state-of-the-art vision transformers. We believe that VisionLLaMA can serve as a strong new baseline model for vision understanding and generation.

Vision-language (VL) models have shown to be very effective in a variety of object detection tasks by utilizing weakly supervised image-text pairs from the web. However, these models exhibit a limited understanding of complex compositions of visual objects (e.g., attributes, shapes, and their relations), resulting in a significant performance drop given complex and diverse language queries. While conventional methods try to enhance VL models through the use of hard negative synthetic augmentation on the text domain, their effectiveness remains restricted without densely paired image-text augmentation. In this paper, we introduce a structured synthetic data generation approach to improve the compositional understanding of VL models for language-based object detection, which generates densely paired positive and negative triplets (object, text descriptions, bounding boxes) in both image and text domains. In addition, in order to train VL models effectively, we propose a new compositional contrastive learning formulation that discovers semantics and structures in complex descriptions from synthetic triplets. As a result, VL models trained with our synthetic data generation exhibit a significant performance boost in the Omnilabel benchmark by up to +5AP and the D3 benchmark by +6.9AP upon existing baselines.

While existing image-text alignment models reach high quality binary assessments, they fall short of pinpointing the exact source of misalignment. In this paper, we present a method to provide detailed textual and visual explanation of detected misalignments between text-image pairs. We leverage large language models and visual grounding models to automatically construct a training set that holds plausible misaligned captions for a given image and corresponding textual explanations and visual indicators. We also publish a new human curated test set comprising ground-truth textual and visual misalignment annotations. Empirical results show that fine-tuning vision language models on our training set enables them to articulate misalignments and visually indicate them within images, outperforming strong baselines both on the binary alignment classification and the explanation generation tasks.

Effectively aligning with human judgment when evaluating machine-generated image captions represents a complex yet intriguing challenge. Existing evaluation metrics like CIDEr or CLIP-Score fall short in this regard as they do not take into account the corresponding image or lack the capability of encoding fine-grained details and penalizing hallucinations. To overcome these issues, in this paper, we propose BRIDGE, a new learnable and reference-free image captioning metric that employs a novel module to map visual features into dense vectors and integrates them into multi-modal pseudo-captions which are built during the evaluation process. This approach results in a multimodal metric that properly incorporates information from the input image without relying on reference captions, bridging the gap between human judgment and machine-generated image captions. Experiments spanning several datasets demonstrate that our proposal achieves state-of-the-art results compared to existing reference-free evaluation scores.

Contextualized Image Captioning (CIC) evolves traditional image captioning into a more complex domain, necessitating the ability for multimodal reasoning. It aims to generate image captions given specific contextual information. This paper further introduces a novel domain of Controllable Contextualized Image Captioning (Ctrl-CIC). Unlike CIC, which solely relies on broad context, Ctrl-CIC accentuates a user-defined highlight, compelling the model to tailor captions that resonate with the highlighted aspects of the context. We present two approaches, Prompting-based Controller (P-Ctrl) and Recalibration-based Controller (R-Ctrl), to generate focused captions. P-Ctrl conditions the model generation on highlight by prepending captions with highlight-driven prefixes, whereas R-Ctrl tunes the model to selectively recalibrate the encoder embeddings for highlighted tokens. Additionally, we design a GPT-4V empowered evaluator to assess the quality of the controlled captions alongside standard assessment methods. Extensive experimental results demonstrate the efficient and effective controllability of our method, charting a new direction in achieving user-adaptive image captioning. Code will be released upon acceptance.

Contrastive vision-language models, such as CLIP, have garnered considerable attention for various dowmsteam tasks, mainly due to the remarkable ability of the learned features for generalization. However, the features they learned often blend content and style information, which somewhat limits their generalization capabilities under distribution shifts. To address this limitation, we adopt a causal generative perspective for multimodal data and propose contrastive learning with data augmentation to disentangle content features from the original representations. To achieve this, we begins with exploring image augmentation techniques and develop a method to seamlessly integrate them into pre-trained CLIP-like models to extract pure content features. Taking a step further, recognizing the inherent semantic richness and logical structure of text data, we explore the use of text augmentation to isolate latent content from style features. This enables CLIP-like model's encoders to concentrate on latent content information, refining the learned representations by pre-trained CLIP-like models. Our extensive experiments across diverse datasets demonstrate significant improvements in zero-shot and few-shot classification tasks, alongside enhanced robustness to various perturbations. These results underscore the effectiveness of our proposed methods in refining vision-language representations and advancing the state-of-the-art in multimodal learning.

We address the challenges inherent in sketch-based image retrieval (SBIR) across various settings, including zero-shot SBIR, generalized zero-shot SBIR, and fine-grained zero-shot SBIR, by leveraging the vision-language foundation model, CLIP. While recent endeavors have employed CLIP to enhance SBIR, these approaches predominantly follow uni-modal prompt processing and overlook to fully exploit CLIP's integrated visual and textual capabilities. To bridge this gap, we introduce \textsc{SpLIP}, a novel multi-modal prompt learning scheme designed to operate effectively with frozen CLIP backbones. We diverge from existing multi-modal prompting methods that either treat visual and textual prompts independently or integrate them in a limited fashion, leading to suboptimal generalization. \textsc{SpLIP} implements a bi-directional prompt-sharing strategy that enables mutual knowledge exchange between CLIP's visual and textual encoders, fostering a more cohesive and synergistic prompt processing mechanism that significantly reduces the semantic gap between the sketch and photo embeddings. In addition to pioneering multi-modal prompt learning, we propose two innovative strategies for further refining the embedding space. The first is an adaptive margin generation for the sketch-photo triplet loss, regulated by CLIP's class textual embeddings. The second introduces a novel task, termed conditional cross-modal jigsaw, aimed at enhancing fine-grained sketch-photo alignment, by focusing on implicitly modelling the …

While exploring visual scenes, humans' scanpaths are driven by their underlying attention processes. Understanding visual scanpaths is essential for various applications. Traditional scanpath models predict the where and when of gaze shifts without providing explanations, creating a gap in understanding the rationale behind fixations. To bridge this gap, we introduce GazeXplain, a novel study of visual scanpath prediction and explanation. This involves annotating natural-language explanations for fixations across eye-tracking datasets and proposing a general model with an attention-language decoder that jointly predicts scanpaths and generates explanations. It integrates a unique semantic alignment mechanism to enhance the consistency between fixations and explanations, alongside a cross-dataset co-training approach for generalization. These novelties present a comprehensive and adaptable solution for explainable human visual scanpath prediction. Extensive experiments on diverse eye-tracking datasets demonstrate the effectiveness of GazeXplain in both scanpath prediction and explanation, offering valuable insights into human visual attention and cognitive processes.

In this paper, we study the problem of Generalized Category Discovery (GCD), which aims to cluster unlabeled data from both known and unknown categories using the knowledge of labeled data from known categories. Current GCD methods rely on only visual cues, which however neglect the multi-modality perceptive nature of human cognitive processes in discovering novel visual categories. To address this, we propose a two-phase TextGCD framework to accomplish multi-modality GCD by exploiting powerful Visual-Language Models. TextGCD mainly includes a retrieval-based text generation (RTG) phase and a cross-modality co-teaching (CCT) phase. First, RTG constructs a visual lexicon using category tags from diverse datasets and attributes from Large Language Models, generating descriptive texts for images in a retrieval manner. Second, CCT leverages disparities between textual and visual modalities to foster mutual learning, thereby enhancing visual GCD. In addition, we design an adaptive class aligning strategy to ensure the alignment of category perceptions between modalities as well as a soft-voting mechanism to integrate multi-modality cues. Experiments on eight datasets show the large superiority of our approach over state-of-the-art methods. Notably, our approach outperforms the best competitor, by 7.7% and 10.8% in All accuracy on ImageNet-1k and CUB, respectively.

We introduce Diff-Tracker, a novel approach for the challenging unsupervised visual tracking task leveraging the pre-trained text-to-image diffusion model. Our main idea is to leverage the rich knowledge encapsulated within the pre-trained diffusion model, such as the understanding of image semantics and structural information, to address unsupervised visual tracking. To this end, we design an initial prompt learner to enable the diffusion model to recognize the tracking target by learning a prompt representing the target. Furthermore, to facilitate dynamic adaptation of the prompt to the target's movements, we propose an online prompt updater. Extensive experiments on five benchmark datasets demonstrate the effectiveness of our proposed method, which also achieves state-of-the-art performance.

Cell tracking is an omnipresent image analysis task in live-cell microscopy. It is similar to multiple object tracking (MOT), however, each frame contains hundreds of similar-looking objects that can divide, making it a challenging problem. Current state-of-the-art approaches follow the tracking-by-detection paradigm, i.e. first all cells are detected per frame and successively linked in a second step to form biologically consistent cell tracks. Linking is commonly solved via discrete optimisation methods, which require manual tuning of hyperparameters for each dataset and are therefore cumbersome to use in practice. Here we propose Trackastra, a general purpose cell tracking approach that uses a simple transformer architecture to directly learn pairwise associations of cells within a temporal window from annotated data. Importantly, unlike existing transformer-based MOT pipelines, our learning architecture also accounts for dividing objects such as cells and allows for accurate tracking even with simple greedy linking, thus removing the requirement for a complex linking step. The proposed architecture operates on the full spatio-temporal context of detections within a time window by avoiding the computational burden of processing dense images. We show that our tracking approach performs on par with or better than highly tuned state-of-the-art cell tracking algorithms for various biological …

Multi-object tracking (MOT) endeavors to precisely estimate the positions and identities of multiple objects over time. The prevailing approach, tracking-by-detection (TbD), first detects objects and then links detections, resulting in a simple yet effective method. However, contemporary detectors may occasionally miss some objects in certain frames, causing trackers to cease tracking prematurely. To tackle this issue, we propose BUSCA, meaning 'to search', a versatile framework compatible with any online TbD system, enhancing its ability to persistently track those objects missed by the detector, primarily due to occlusions. Remarkably, this is accomplished without modifying past tracking results or accessing future frames, i.e., in a fully online manner. BUSCA generates proposals based on neighboring tracks, motion, and learned tokens. Utilizing a decision Transformer that integrates multimodal visual and spatiotemporal information, it addresses the object-proposal association as a multi-choice question-answering task. BUSCA is trained independently of the underlying tracker, solely on synthetic data, without requiring fine-tuning. Through BUSCA, we showcase consistent performance enhancements across five different trackers and establish a new state-of-the-art baseline across three different benchmarks. We will publicly release the code to facilitate its integration into future MOT methods.

The supervision of state-of-the-art multiple object tracking (MOT) methods requires enormous annotation efforts to provide bounding boxes for all frames of all videos, and instance IDs to associate them through time. To this end, we introduce Walker, the first self-supervised tracker that learns from videos with sparse bounding box annotations, and no tracking labels. First, we design a quasi-dense temporal object appearance graph, and propose a novel multi-positive contrastive objective to optimize random walks on the graph and learn instance similarities. Then, we introduce an algorithm to enforce mutually-exclusive connective properties across instances in the graph, optimizing the learned topology for MOT. At inference time, we propose to associate detected instances to tracklets based on the max-likelihood transition state under motion-constrained bi-directional walks. Walker is the first self-supervised tracker to achieve competitive performance on MOT17, DanceTrack, and BDD100K. Remarkably, our proposal outperforms the previous self-supervised trackers even when drastically reducing the annotation requirements by up to 400x.

Compared with visual signals, Inertial Measurement Units (IMUs) placed on human limbs can capture accurate motion signals while being robust to lighting variation and occlusion. While these characteristics are intuitively valuable to help egocentric action recognition, the potential of IMUs remains under-explored. In this work, we present a novel method for action recognition that integrates motion data from body-worn IMUs with egocentric video. Due to the scarcity of labeled multimodal data, we design an MAE-based self-supervised pretraining method, obtaining strong multi-modal representation via modeling the natural correlation between visual and motion signals. To model the complex relation of multiple IMU devices placed across the body, we exploit the collaborative dynamics in multiple IMU devices and propose to embed the relative motion features of human joints into a graph structure. Experiments show our method can achieve state-of-the-art performance on multiple public datasets. The effectiveness of our MAE-based pretraining and graph-based IMU modeling are further validated by experiments in more challenging scenarios, including partially missing IMU devices and video quality corruption, promoting more flexible usages in the real world.

This paper presents a self-supervised temporal video alignment framework which is useful for several fine-grained human activity understanding applications. In contrast with the state-of-the-art method of CASA, where sequences of 3D skeleton coordinates are taken directly as input, our key idea is to use sequences of 2D skeleton heatmaps as input. Unlike CASA which performs self-attention in the temporal domain only, we feed 2D skeleton heatmaps to a video transformer which performs self-attention both in the spatial and temporal domains for extracting effective spatiotemporal and contextual features. In addition, we introduce simple heatmap augmentation techniques based on 2D skeletons for self-supervised learning. Despite the lack of 3D information, our approach achieves not only higher accuracy but also better robustness against missing and noisy keypoints than CASA. Furthermore, extensive evaluations on three public datasets, i.e., Penn Action, IKEA ASM, and H2O, demonstrate that our approach outperforms previous methods in different fine-grained human activity understanding tasks. Finally, fusing 2D skeleton heatmaps with RGB videos yields the state-of-the-art on all metrics and datasets. To our best knowledge, our work is the first to utilize 2D skeleton heatmap inputs and the first to explore multi-modality fusion for temporal video alignment.

Extensive occlusions in real-world scenarios pose challenges to gait recognition due to missing and noisy information, as well as body misalignment in position and scale. We argue that rich dynamic contextual information within a gait sequence inherently possesses occlusion-solving traits: 1) Adjacent frames with gait continuity allow holistic body regions to infer occluded body regions; 2) Gait cycles allow information integration between holistic actions and occluded actions. Therefore, we introduce an action detection perspective where a gait sequence is regarded as a composition of actions. To detect accurate actions under complex occlusion scenarios, we propose an Action Detection Based Mixture of Experts (GaitMoE), consisting of Mixture of Temporal Experts (MTE) and Mixture of Action Experts (MAE). MTE adaptively constructs action anchors by temporal experts and MAE adaptively constructs action proposals from action anchors by action experts. Especially, action detection as a proxy task with gait recognition is an end-to-end joint training only with ID labels. In addition, due to the lack of a unified occluded benchmark, we construct a pioneering Occluded Gait database (OccGait), containing rich occlusion scenarios and annotations of occlusion types. Extensive experiments on OccGait, OccCASIA-B, Gait3D and GREW demonstrate the superior performance of GaitMoE. OccGait is available …

Point-supervised temporal action localization pursues high-accuracy action detection under low-cost data annotation. Despite recent advances, a significant challenge remains: sparse labeling of individual frames leads to semantic ambiguity in determining action boundaries due to the lack of continuity in the highly sparse point-supervision scheme. We propose a Stepwise Multi-grained Boundary Detector(SMBD), which is comprised of a Background Anchor Generator(BAG) and a Dual Boundary Detector(DBD) to provide fine-grained supervision. Specifically, for each epoch in the training process, BAG computes the optimal background snippet between each pair of adjacent action labels, which we term Background Anchor. Subsequently, DBD leverages the background anchor and the action labels to locate the action boundaries from the perspectives of detecting action changes and scene changes. Then, the corresponding labels can be assigned to each side of the boundaries, with the boundaries continuously updated throughout the training process. Consequently, the proposed SMBD could ensure that more snippets contribute to the training process. Extensive experiments on the THUMOS’14, GTEA and BEOID datasets demonstrate that the proposed method outperforms existing state-of-the-art methods.

Event cameras excel in capturing high-contrast scenes and dynamic objects, offering a significant advantage over traditional frame-based cameras. Despite active research into leveraging event cameras for semantic segmentation, generating pixel-wise dense semantic maps for such challenging scenarios remains labor-intensive. As a remedy, we present EV-WSSS: a novel weakly supervised approach for event-based semantic segmentation that utilizes sparse point annotations. To fully leverage the temporal characteristics of event data, the proposed framework performs asymmetric dual-student learning between 1) the original forward event data and 2) the longer reversed event data, which contain complementary information from the past and the future, respectively. Besides, to mitigate the challenges posed by sparse supervision, we propose feature-level contrastive learning based on class-wise prototypes, carefully aggregated at both spatial region and sample levels. Additionally, we further excavate the potential of our dual-student learning model by exchanging prototypes between the two learning paths, thereby harnessing their complementary strengths. With extensive experiments on various datasets, including DSEC Night-Point with sparse point annotations newly provided by this paper, the proposed method achieves substantial segmentation results even without relying on pixel-level dense ground truths. The code and dataset will be published soon.

Multi-modal test-time adaptation (MM-TTA) is proposed to adapt models to an unlabeled target domain by leveraging the complementary multi-modal inputs in an online manner. Previous MM-TTA methods rely on predictions of cross-modal information in each input frame, while they ignore the fact that predictions of geometric neighborhoods within consecutive frames are highly correlated, leading to unstable predictions across time. To fulfill this gap, we propose ReLiable Spatial-temporal Voxels (Latte), an MM-TTA method that leverages reliable cross-modal spatial-temporal correspondences for multi-modal 3D segmentation. Motivated by the fact that reliable predictions should be consistent with their spatial-temporal correspondences, Latte aggregates consecutive frames in a slide window manner and constructs ST voxel to capture temporally local prediction consistency for each modality. After filtering out ST voxels with high ST entropy, Latte conducts cross-modal learning for each point and pixel by attending to those with reliable and consistent predictions among both spatial and temporal neighborhoods. Experimental results show that Latte achieves state-of-the-art performance on three different MM-TTA benchmarks compared to previous MM-TTA or TTA methods.

This work aims to address an advanced keypoint detection problem: how to accurately detect any keypoints in complex real-world scenarios, which involves massive, messy, and open-ended objects as well as their associated keypoints definitions. Current high-performance keypoint detectors often fail to tackle this problem due to their two-stage schemes, under-explored prompt designs, and limited training data. To bridge the gap, we propose the X-Pose, a novel end-to-end framework with multi-modal (i.e., visual, textual or their combinations) prompts to detect multi-object keypoints for any articulated (e.g., human and animal), rigid, and soft objects within a given image. Moreover, we introduce a large-scale dataset called UniKPT, which unifies 13 keypoint detection datasets with 338 keypoints across 1,237 categories over 400K instances. Training with UniKPT, X-Pose effectively aligns text-to-keypoint and image-to-keypoint due to the mutual enhancement of multi-modal prompts based on cross-modality contrastive learning. Our experimental results demonstrate that X-Pose achieves notable improvements of 27.7 AP, 6.44 PCK and 7.0 AP compared to state-of-the-art non-promptable, visual prompt-based, and textual prompt-based methods in each respective fair setting. More importantly, the in-the-wild test demonstrates X-Pose's strong fine-grained keypoint localization and generalization abilities across image styles, object categories, and poses, paving a new path to mutli-object …

Are vision-language models (VLMs) open-set models because they are trained on internet-scale datasets? We answer this question with a clear no -- VLMs introduce closed-set assumptions via their finite query set, making them vulnerable to open-set conditions. We systematically evaluate VLMs for open-set recognition and find they frequently misclassify objects not contained in their query set, leading to alarmingly low precision when tuned for high recall and vice versa. We show that naively increasing the size of the query set to contain more and more classes does not mitigate this problem, but instead causes diminishing task performance and open-set performance. We establish a revised definition of the open-set problem for the age of VLMs, define a new benchmark and evaluation protocol to facilitate standardised evaluation and research in this important area, and evaluate promising baseline approaches based on predictive uncertainty and dedicated negative embeddings on a range of VLM classifiers and object detectors.

In panoptic scene graph generation (PSGG), models retrieve interactions between objects in an image which are grounded by panoptic segmentation masks. Previous evaluations on panoptic scene graphs have been subject to an erroneous evaluation protocol where multiple masks for the same object can lead to multiple relation distributions per mask-mask pair. This can be exploited to increase the final score. We correct this flaw and provide a fair ranking over a wide range of existing PSGG models. The observed scores for existing methods increase by up to 7.4 mR@50 for all two-stage methods, while dropping by up to 19.3 mR@50 for all one-stage methods, highlighting the importance of a correct evaluation. Contrary to recent publications, we show that existing two-stage methods are competitive to one-stage methods. Building on this, we introduce the Decoupled SceneFormer (DSFormer), a novel two-stage model that outperforms all existing scene graph models by a large margin of +11 mR@50 and +10 mNgR@50 on the corrected evaluation, thus setting a new SOTA. As a core design principle, DSFormer encodes subject and object masks directly into feature space.

Open-vocabulary 3D object detection (OV-3DDET) is a challenging task aimed at locating and recognizing objects in a 3D scene, encompassing both seen and previously unseen categories. Unlike in the vision and language domain where abundant training data is available to train generalized models, 3D detection models suffer from the scarcity of training data. Despite this challenge, the flourishing field of vision-language models (VLMs) offers valuable insights that can guide the learning process for OV-3DDET. While some efforts have been made to incorporate VLMs into OV-3DDET learning, existing methods often fall short in establishing a comprehensive association between 3D detectors and VLMs. In this paper, we investigate the utilization of VLMs for the task of open-vocabulary 3D detection. We use a vision model to guide novel class discovery in 3D scenes, and hierarchically align the 3D feature and vision-language feature space. Specifically, we employ an off-the-shelf 2D detector to seed and select novel 3D objects. The discovered novel objects are then stored for retraining the 3D detector. Finally, we align the 3D feature space with the vision-language feature space using a pre-trained vision-language model at the instance, category, and scene levels. Through extensive experimentation, we demonstrate substantial improvements in accuracy and …

Motivated by the Parameter-Efficient Fine-Tuning (PEFT) in large language models, we propose LoRAT, a method that unveils the power of larger Vision Transformers (ViT) for tracking within laboratory-level resources. The essence of our work lies in adapting LoRA, a technique that fine-tunes a small subset of model parameters without adding inference latency, to the domain of visual tracking. However, unique challenges and potential domain gaps make this transfer not as easy as the first intuition. Firstly, a transformer-based tracker constructs unshared position embedding for template and search image. This poses a challenge for the transfer of LoRA, usually requiring consistency in the design when applied to the pre-trained backbone, to downstream tasks. Secondly, the inductive bias inherent in convolutional heads diminishes the effectiveness of parameter-efficient fine-tuning in tracking models. To overcome these limitations, we first decouple the position embeddings in transformer-based trackers into shared spatial ones and independent type ones. The shared embeddings, which describe the absolute coordinates of multi-resolution images (namely, the template and search images), are inherited from the pre-trained backbones. In contrast, the independent embeddings indicate the sources of each token and are learned from scratch. Furthermore, we design an anchor-free head solely based on a multilayer …

In computer vision, it is well-known that a lack of data diversity will impair model performance. In this study, we address the challenges of enhancing the dataset diversity problem in order to benefit various downstream tasks such as object detection and instance segmentation. We propose a simple yet effective data augmentation approach by leveraging advancements in generative models, specifically text-to-image synthesis technologies like Stable Diffusion. Our method focuses on generating variations of labeled real images, utilizing generative object and background augmentation via inpainting to augment existing training data without the need for additional annotations. We find that background augmentation, in particular, significantly improves the models' robustness and generalization capabilities. We also investigate how to adjust the prompt and mask to ensure the generated content do not violate the existing annotations. The efficacy of our augmentation techniques is validated through comprehensive evaluations of the COCO dataset and several other key object detection benchmarks, demonstrating notable enhancements in model performance across diverse scenarios. This approach offers a promising solution to the challenges of dataset enhancement, contributing to the development of more accurate and robust computer vision models.

In this work, we introduce OpenIns3D, a new framework for 3D open-vocabulary scene understanding at instance level. Unlike all existing methods, the proposed pipeline requires no well-aligned images as input and works effectively on a wide range of scenarios. The OpenIns3D framework employs a "Mask-Snap-Lookup" scheme, where the "Mask" module learns class-agnostic mask proposals in 3D point clouds, the "Snap" module generates synthetic scene-level images at multiple scales and leverages 2D vision language models to extract interesting objects, and the "Lookup" module searches through the outcomes of "Snap" to assign category names to the proposed masks. This approach, free from 2D input requirements yet simple and flexible, achieves state-of-the-art performance across a wide range of 3D open-vocabulary tasks, including recognition, object detection, and instance segmentation, on both indoor and outdoor datasets. Moreover, OpenIns3D facilitates effortless switching between different 2D detectors without requiring retraining. When integrated with powerful 2D open-world models, it achieves excellent results in scene understanding tasks. Furthermore, when combined with LLM-powered 2D models, OpenIns3D exhibits a remarkable capability to comprehend and process highly complex text queries that demand intricate reasoning and real-world knowledge.

Traditional 3D scene understanding techniques rely on supervised learning from densely annotated 3D datasets. However, the collection and annotation of 3D data is expensive and tedious, which leads to the scarcity of labeled training data. In this paper, we investigate the use of diffusion models which are pre-trained on large-scale image-caption pairs for open-vocabulary 3D scene understanding. We propose a novel method, namely Diff2Scene, leverages frozen representations from text-image discriminative and generative models, along with salient-aware and geometric-aware masks, for open-vocabulary scene understanding. Diff2Scene gets rid of any labeled 3D data and effectively identifies objects, appearances, materials, locations and their compositions in 3D scenes using a single model. We show that it outperforms competitive baselines and achieves significant improvements over state-of-the-art methods in open-vocabulary 3D semantic segmentation tasks. In particular, Diff2Scene improves the state-of-the-art method on ScanNet200 by 12%.

The attention module is the key component in Transformers. While the global attention mechanism offers high expressiveness, its excessive computational cost restricts its applicability in various scenarios. In this paper, we propose a novel attention paradigm, Agent Attention, to strike a favorable balance between computational efficiency and representation power. Specifically, the Agent Attention, denoted as a quadruple (Q, A, K, V), introduces an additional set of agent tokens A into the conventional attention module. The agent tokens first act as the agent for the query tokens Q to aggregate information from K and V, and then broadcast the information back to Q. Given the number of agent tokens can be designed to be much smaller than the number of query tokens, agent attention is significantly more efficient than the widely adopted Softmax attention, while preserving global context modelling capability. Interestingly, we show that the proposed agent attention is equivalent to a generalized form of linear attention. Therefore, agent attention seamlessly integrates the powerful Softmax attention and the highly efficient linear attention. Extensive experiments demonstrate the effectiveness of agent attention with various vision Transformers and across diverse vision tasks, including image classification, object detection, semantic segmentation and image generation. Notably, agent …

Transcription-only Supervised Text Spotting aims to learn text spotters relying only on transcriptions but no text boundaries for supervision, thus eliminating expensive boundary annotation. The crux of this task lies in locating each transcription in scene text images without location annotations. In this work, we formulate this challenging problem as a Weakly Supervised Cross-modality Contrastive Learning problem, and design a simple yet effective model dubbed WeCromCL that is able to detect each transcription in a scene image in a weakly supervised manner. Unlike typical methods for cross-modality contrastive learning that focus on modeling the holistic semantic correlation between an entire image and a text description, our WeCromCL conducts atomistic contrastive learning to model the character-wise appearance consistency between a text transcription and its correlated region in a scene image to detect an anchor point for the transcription in a weakly supervised manner. The detected anchor points by WeCromCL are further used as pseudo location labels to guide the learning of text spotting. Extensive experiments on four challenging benchmarks demonstrate the superior performance of our model over other methods. Code will be released.

We present Agglomerative Token Clustering (ATC), a novel token merging method that consistently outperforms previous token merging and pruning methods across image classification, image synthesis, and object detection & segmentation tasks. ATC merges clusters through bottom-up hierarchical clustering, without the introduction of extra learnable parameters. We find that ATC achieves state-of-the-art performance across all tasks, and can even perform on par as the prior state-of-the-art when applied off-the-shelf, i.e. without fine-tuning. ATC is particularly effective when applied with low keep rates, where only a small fraction of tokens are kept and retaining task performance is especially difficult.

We present an Encoder-Decoder Attention Transformer, EDAFormer, which consists of the Embedding-Free Transformer (EFT) encoder and the all-attention decoder leveraging our Embedding-Free Attention (EFA) structure. The proposed EFA is a novel global context modeling mechanism that focuses on functioning the global non-linearity, not the specific roles of the query, key and value. For the decoder, we explore the optimized structure for considering the globality, which can improve the semantic segmentation performance. In addition, we propose a novel Inference Spatial Reduction (ISR) method for the computational efficiency. Different from the previous spatial reduction attention methods, our ISR method reduces the key-value resolution at the inference, which can reduce the computation-performance trade-off gap for the efficient semantic segmentation. Our EDAFormer shows the state-of-the-art performance with the efficient computation compared to the existing transformer-based semantic segmentation models in three public benchmarks, including ADE20K, Cityscapes and COCO-Stuff. Furthermore, our ISR method reduces the computational cost by up to 61% with minimal mIoU performance degradation on Cityscapes dataset.

3D weakly supervised semantic segmentation aims to learn semantic segmentation without using dense annotations. Previous methods mostly use Class Activation Map to solve this challenge. In such a paradigm, the model is supervised given the scene-level or subcloud-level labels, however, remaining less-explored in the potential textually semantic information from the category labels. In this paper, we propose 3DSS-VLG, a weakly supervised approach for 3D Semantic Segmentation with 2D Vision-Language Guidance, an alternative approach that a 3D model predicts dense-embedding for each point which is co-embedded with both the aligned image and text spaces from the 2D vision-language model. Specifically, our method exploits the superior generalization ability of the 2D vision-language models and proposes Embeddings Soft-Guidance Stage to utilize it to implicitly align 3D embeddings and text embeddings. Moreover, we introduce Embeddings Specialization Stage to purify the feature representation with the help of given scene-level label, specifying better feature supervised by the corresponding text embedding. Thus, the 3D model is able to gain the informative supervisions both from the image embedding and text embedding, leading to competitive segmentation performances. To the best of our knowledge, this is the first work to investigate 3D weakly supervised semantic segmentation by using the textual …

Hierarchical segmentation entails creating segmentations at varying levels of granularity. We introduce the first hierarchical semantic segmentation dataset with subpart annotations for natural images, which we call SPIN (SubPartImageNet). We also introduce two novel evaluation metrics to evaluate how well algorithms capture spatial and semantic relationships across hierarchical levels. We benchmark modern models across three different tasks and analyze their strengths and weaknesses across objects, parts, and subparts. To facilitate community-wide progress, we publicly release our dataset at https://joshmyersdean.github.io/spin/index.html.

RGB-Thermal (RGB-T) semantic segmentation is an important research branch of multi-modal image segmentation. The current RGB-T semantic segmentation methods generally have two unsolved and typical shortcomings. First, they do not have the open-vocabulary recognition ability, which significantly limits their application scenarios. Second, when fusing RGB and thermal images, they often need to design complex fusion network structures, which usually results in low network training efficiency. We present OpenRSS, the Open-vocabulary RGB-T Semantic Segmentation method, to solve these two disadvantages. To our knowledge, OpenRSS is the first RGB-T semantic segmentation method with open-vocabulary segmentation capability. OpenRSS modifies the basic segmentation model SAM for RGB-T semantic segmentation by adding the proposed thermal information prompt module and dynamic low-rank adaptation strategy to SAM. These designs effectively fuse the RGB and thermal information, but with much fewer trainable parameters than other methods. OpenRSS achieves the open-vocabulary capability by jointly utilizing the vision-language model CLIP and the modified SAM. Through extensive experiments, OpenRSS demonstrates its effective open-vocabulary semantic segmentation ability on RGB-T images. It outperforms other state-of-the-art RGB open-vocabulary semantic segmentation methods on multiple RGB-T semantic segmentation benchmarks: +12.1% mIoU on the MFNet dataset, +18.4% mIoU on the MCubeS dataset, and +21.4% mIoU on the …

We introduce PartSTAD, a method designed for the task adaptation of 2D-to-3D segmentation lifting. Recent studies have highlighted the advantages of utilizing 2D segmentation models to achieve high-quality 3D segmentation through few-shot adaptation. However, previous approaches have focused on adapting 2D segmentation models for domain shift to rendered images and synthetic text descriptions, rather than optimizing the model specifically for 3D segmentation. Our proposed task adaptation method finetunes a 2D bounding box prediction model with an objective function for 3D segmentation. We introduce weights for 2D bounding boxes for adaptive merging and learn the weights using a small additional neural network. Additionally, we incorporate SAM, a foreground segmentation model on a bounding box, to improve the boundaries of 2D segments and consequently those of 3D segmentation. Our experiments on the PartNet-Mobility dataset show significant improvements with our task adaptation approach, achieving a 7.0%p increase in mIoU and a 5.2%p improvement in mAP@50 for semantic and instance segmentation compared to the SotA few-shot 3D segmentation model. The code will be released publicly later.

The CLIP and Segment Anything Model (SAM) are remarkable vision foundation models (VFMs). SAM excels in segmentation tasks across diverse domains, while CLIP is renowned for its zero-shot recognition capabilities. This paper presents an in-depth exploration of integrating these two models into a unified framework. Specifically, we introduce the Open-Vocabulary SAM, a SAM-inspired model designed for simultaneous interactive segmentation and recognition, leveraging two unique knowledge transfer modules: SAM2CLIP and CLIP2SAM. The former adapts SAM's knowledge into the CLIP via distillation and learnable transformer adapters, while the latter transfers CLIP knowledge into SAM, enhancing its recognition capabilities. Extensive experiments on various datasets and detectors show the effectiveness of Open-Vocabulary SAM in both segmentation and recognition tasks, significantly outperforming the na\"{i}ve baselines of simply combining SAM and CLIP. Furthermore, aided with image classification data training, our method can segment and recognize approximately 22,000 classes. The code and models will be publicly available for further research.

Robust semantic segmentation under adverse conditions is of great importance in real-world applications. To address this challenging task in practical scenarios where labeled normal condition images are not accessible in training, we propose FREST, a novel feature restoration framework for source-free domain adaptation (SFDA) of semantic segmentation to adverse conditions. FREST alternates two steps: (1) learning the condition embedding space that only separates the condition information from the features and (2) restoring features of adverse condition images on the learned condition embedding space. By alternating these two steps, FREST gradually restores features where the effect of adverse conditions is reduced. FREST achieved a state of the art on two public benchmarks (i.e., ACDC and RobotCar) for SFDA under adverse conditions. Moreover, it shows superior generalization ability on unseen datasets.

The labor-intensive labeling for semantic segmentation has spurred the emergence of Unsupervised Semantic Segmentation. Recent studies utilize patch-wise contrastive learning based on features from image-level self-supervised pretrained models. However, relying solely on similarity-based supervision from image-level pretrained models often leads to unreliable guidance due to insufficient patch-level semantic representations. To address this, we propose Progressive Proxy Anchor Propagation (PPAP) strategy. This method gradually identifies more trustworthy positives of each anchor by relocating its proxy to densely populated regions of semantically similar samples. Specifically, we initially establish a tight boundary to gather a few reliable positive samples around each anchor. Then, considering the distribution of positive samples, we relocate the proxy anchor towards areas with a higher concentration of positives and adjust the positiveness boundary based on the propagation degree of the proxy anchor. In addition, there might exist ambiguous regions where positive and negative samples coexist near the positiveness boundary. Therefore, to further ensure the reliability of the negative set, we define an instance-wise ambiguous zone and exclude samples in such regions from the negative set. Our state-of-the-art performances on various datasets validate the effectiveness of the proposed method for Unsupervised Semantic Segmentation.

Class incremental semantic segmentation aims to preserve old knowledge while learning new tasks, however, impeded by catastrophic forgetting and background shift issues. Prior works indicate the pivotal importance of initializing new classifiers and mainly focus on transferring knowledge from the background classifier or preparing classifiers for future classes, neglecting the alignment and variance of new classifiers. In this paper, we propose a new classifier pre-tuning (NeST) method applied before the formal training process, learning a transformation from old classifiers to generate new classifiers for initialization rather than directly tuning the parameters of new classifiers. Our method can make new classifiers align with the backbone and adapt the new data, benefiting both the stability and plasticity of the model. Besides, we design a strategy considering the cross-task class similarity to initialize matrices used in the transformation. Experiments on Pascal VOC 2012 and ADE20K datasets show that the proposed strategy can significantly improve the performance of previous methods.

Machine learning models are vulnerable to tiny adversarial input perturbations optimized to cause a very large output error. To measure this vulnerability, we need reliable methods that can find such adversarial perturbations. For image classification models, evaluation methodologies have emerged that have stood the test of time. However, we argue that in the area of semantic segmentation, a good approximation of the sensitivity to adversarial perturbations requires significantly more effort than what is currently considered satisfactory. To support this claim, we re-evaluate a number of well-known robust segmentation models in an extensive empirical study. We propose new attacks and combine them with the strongest attacks available in the literature. We also analyze the sensitivity of the models in fine detail. The results indicate that most of the state-of-the-art models have a dramatically larger sensitivity to adversarial perturbations than previously reported. We also demonstrate a size-bias: small objects are often more easily attacked, even if the large objects are robust, a phenomenon not revealed by current evaluation metrics. Our results also demonstrate that a diverse set of strong attacks is necessary, because different models are often vulnerable to different attacks.

Existing generalized few-shot 3D segmentation (GFS3DS) methods typically prioritize enhancing the training of base-class prototypes while neglecting the rich semantic information within background regions for future novel classes. We introduce a novel GFS3DS learner that strategically leverages background context to improve both base prototype training and few-shot adaptability. Our method employs foundation models to extract semantic features from background points and grounds on text embeddings to cluster background points into pseudo-classes. This approach facilitates clearer base/novel class differentiation and generates pseudo prototypes that effectively mimic novel support samples. Comprehensive experiments on S3DIS and ScanNet datasets demonstrate the state-of-the-art performance of our method in both 1-shot and 5-shot tasks. Our approach significantly advances GFS3DS by unlocking the potential of background context, offering a promising avenue for broader applications. The GitHub repository of our implementation will be released upon publication.

Our paper introduces a novel two-stage self-supervised approach for detecting co-occurring salient objects (CoSOD) in image groups without requiring segmentation annotations. Unlike existing unsupervised methods that rely solely on patch-level information (e.g. clustering patch descriptors) or on computation heavy off-the-shelf components for CoSOD, our lightweight model leverages feature correspondences at both patch and region levels, significantly improving prediction performance. In the first stage, we train a self-supervised network that detects co-salient regions by computing local patch-level feature correspondences across images. We obtain the segmentation predictions using confidence-based adaptive thresholding. In the next stage, we refine these intermediate segmentations by eliminating the detected regions (within each image) whose averaged feature representations are dissimilar to the foreground feature representation averaged across all the cross-attention maps (from the previous stage). Extensive experiments on three CoSOD benchmark datasets show that our self-supervised model outperforms the corresponding state-of-the-art models by a huge margin (e.g. on the CoCA dataset, our model has a 13.7% F-measure gain over the SOTA unsupervised CoSOD model). Notably, our self-supervised model also outperforms several recent fully supervised CoSOD models on the three test datasets (\eg, on the CoCA dataset, our model has a 4.6% F-measure gain over a recent supervised CoSOD …

With the emergent attentive property of self-supervised Vision Transformer (ViT), Normalized Cuts (NCut) has resurfaced as a powerful tool for unsupervised dense prediction. However, the pre-trained ViT backbone (e.g., DINO) is frozen in existing methods, which makes the feature extractor suboptimal for dense prediction tasks. In this paper, we propose using Differentiable Normalized Cuts for self-supervised dense feature learning that can improve the dense prediction capability of existing pre-trained models. First, we review an efficient gradient formulation for the classical NCut algorithm. This formulation only leverages matrices computed and stored in the forward pass, making the backward pass highly efficient. Second, with NCut gradients in hand, we design a self-supervised dense feature learning architecture to finetune pre-trained models. Given two random augmented crops of an image, the architecture performs RoIAlign and NCut to generate two foreground masks of their overlapping region. Last, we propose a mask-consistency loss to back-propagate through NCut and RoIAlign for model training. Experiments show that our framework generalizes to various pre-training methods (DINO, MoCo and MAE), network configurations (ResNet, ViT-S and ViT-B), and tasks (unsupervised saliency detection, object discovery and semantic correspondence). Moreover, we achieved state-of-the-art results on unsupervised dense prediction benchmarks.

The existing crowd counting models require extensive training data, which is time-consuming to annotate. To tackle this issue, we propose a simple yet effective crowd counting method by utilizing the Segment-Everything-Everywhere Model (SEEM), an adaptation of the Segmentation Anything Model (SAM), to generate pseudo-labels for training crowd counting models. However, our initial investigation reveals that SEEM's performance in dense crowd scenes is limited, primarily due to the omission of many persons in high-density areas. To overcome this limitation, we propose an adaptive resolution SEEM to handle the scale variations, occlusions, and overlapping of people within crowd scenes. Alongside this, we introduce a robust localization method, based on Gaussian Mixture Models, for predicting the head positions in the predicted people masks. Given the mask and point pseudo-labels, we propose a robust loss function, which is designed to exclude uncertain regions based on SEEM's predictions, thereby enhancing the training process of the counting networks. Finally, we propose an iterative method for generating pseudo-labels. This method aims at improving the quality of the segmentation masks by identifying more tiny persons in high-density regions, which are often missed in the first pseudo-labeling stage. Overall, our proposed method achieves the best unsupervised performance in crowd …

Acquiring fine-grained object detection annotations in unconstrained images is time-consuming, expensive, and prone to noise, especially in crowdsourcing scenarios. Most prior object detection methods assume accurate annotations; A few recent works have studied object detection with noisy crowdsourced annotations, with evaluation on distinct synthetic crowdsourced datasets of varying setups under artificial assumptions. To address these algorithmic limitations and evaluation inconsistency, we first propose a novel Bayesian Detector Combination (BDC) framework to more effectively train object detectors with noisy crowdsourced annotations, with the unique ability of automatically inferring the annotators' label qualities. Unlike previous approaches, BDC is model-agnostic, requires no prior knowledge of the annotators' skill level, and seamlessly integrates with existing object detection models. Due to the scarcity of real-world crowdsourced datasets, we introduce large synthetic datasets by simulating varying crowdsourcing scenarios. This allows consistent evaluation of different models at scale. Extensive experiments on both real and synthetic crowdsourced datasets show that BDC outperforms existing state-of-the-art methods, demonstrating its superiority in leveraging crowdsourced data for object detection.

In incremental object detection, knowledge distillation has been proven to be an effective way to alleviate catastrophic forgetting. However, previous works focused on preserving the knowledge of old models, ignoring that images could simultaneously contain categories from past, present, and future stages. The co-occurrence of objects makes the optimization objectives inconsistent across different stages since the definition for foreground objects differs across various stages, greatly limiting the model's performance. To overcome this problem, we propose a method called ``Bridge Past and Future'' (BPF), which aligns models across stages, ensuring consistent optimization directions. In addition, we propose a novel Distillation with Future (DwF) loss, fully leveraging the background probability to mitigate the forgetting of old classes while ensuring a high level of adaptability in learning new classes. Extensive experiments are conducted on both Pascal VOC and MS COCO benchmarks. Without memory, BPF outperforms current state-of-the-art methods under various settings.

Domain Adaptation (DA) has been widely explored and made significant progress on cross-domain 3D tasks recently. Despite being effective, existing works fail to deal with rapidly changing domains due to the unpredictable test time scenarios and meanwhile fast response time requirement. Thus, we explore a new task named test-time domain adaptive 3D object detection and propose RegTTA3D, a pseudo-label-based test-time adaptative 3D object detection method. By investigating the factor that limits the detection accuracy, we find that regression is essential in this task. To make better regression, we first design a noise-consistency pseudo-label generation process to filter pseudo-labels with instability under noise interference and obtain reliable pseudo-labels. Then, confidence-guided regression refinement is introduced, which uses the box regression results of high-confidence boxes to supervise boxes with relatively low confidence, further making the predicted box size gradually approach the distribution of the target domain. Finally, to better update the regression layer and alleviate the class-imbalance issue, a class-balance EMA updating strategy is proposed. Experimental results on multiple cross-domain scenarios including cross-beam, crosslocation, and cross-weather demonstrate that Reg-TTA3D can achieve comparable or even better performance compared to unsupervised domain adaptation works by only updating less than 0.1% parameters within less than 1% …

Detection pre-training methods for the DETR series detector have been extensively studied in natural scenes, i.e., DETReg. However, the detection pre-training remains unexplored in remote sensing scenes. In existing pre-training methods, alignment between object embeddings extracted from a pre-trained backbone and detector features is significant. However, due to differences in feature extraction ways, a pronounced feature discrepancy still exists and hinders the pre-training performance. The remote sensing images with complex environments and more densely distributed objects exacerbate the discrepancy. In this work, we propose a novel Mutually optimizing pre-training framework for remote sensing object Detection, dubbed as MutDet. In MutDet, we propose a systemic solution against this challenge. Firstly, we propose a mutual enhancement module, which fuses the object embeddings and detector features bidirectionally in the last encoder layer, enhancing their information interaction. Secondly, contrastive alignment loss is employed to guide this alignment process softly and simultaneously enhances detector features' discriminativity. Finally, we design an auxiliary siamese head to mitigate the task gap arising from the introduction of enhancement module. Comprehensive experiments on various settings show new state-of-the-art transfer performance. The improvement is particularly pronounced when data quantity is limited. When using 10 % of the DIOR-R data, MutDet improves …

The recent Segment Anything Model (SAM) is a significant advancement in natural image segmentation, exhibiting potent zero-shot performance suitable for various downstream image segmentation tasks. However, directly utilizing the pretrained SAM for Infrared Small Target Detection (IRSTD) task falls short in achieving satisfying performance due to a notable domain gap between natural and infrared images. Unlike a visible light camera, a thermal imager reveals an object's temperature distribution by capturing infrared radiation. Small targets often show a subtle temperature transition at the object's boundaries. To address this issue, we propose the IRSAM model for IRSTD, which improves SAM's encoder-decoder architecture to learn better feature representation of infrared small objects. Specifically, we design a Perona-Malik diffusion (PMD)-based block and incorporate it into multiple levels of SAM's encoder to help it capture essential structural features while suppressing noise. Additionally, we devise a Granularity-Aware Decoder (GAD) to fuse the multi-granularity feature from the encoder to capture structural information that may be lost in long-distance modeling. Extensive experiments on the public datasets, including NUAA-SIRST, NUDT-SIRST, and IRSTD-1K, validate the design choice of IRSAM and its significant superiority over representative state-of-the-art methods. The source code will be released.

In digital pathology, segmenting densely distributed objects like glands and nuclei is crucial for downstream analysis. Since detailed pixel-wise annotations are very time-consuming, we need semi-supervised segmentation methods that can learn from unlabeled images. Existing semi-supervised methods are often prone to topological errors, e.g., missing or incorrectly merged/separated glands or nuclei. To address this issue, we propose TopoSemiSeg, the first semi-supervised method that learns the topological representation from unlabeled histopathology images. The major challenge is for unlabeled images; we only have predictions carrying noisy topology. To this end, we introduce a noise-aware topological consistency loss to align the representations of a teacher and a student model. By decomposing the topology of the prediction into signal topology and noisy topology, we ensure that the models learn the true topological signals and become robust to noise. Extensive experiments on public histopathology image datasets show the superiority of our method, especially on topology-aware evaluation metrics.

Despite the recent success of domain generalization in medical image segmentation, it is known that precisely conducting voxel-wise annotation for all available source domains remains a huge burden. To combat this challenge, as a new setting, semi-supervised domain generalization has been proposed very recently by leveraging limited labeled data along with abundant unlabeled data collected from multiple medical institutions to enhance model generalization. To achieve promising results in this setting, correctly harnessing unlabeled data while improving the generalization ability to unseen domains plays a critical role simultaneously. In this work, we observe that the domain shifts between medical institutions (i.e., source domains) cause disparate feature statistics, which significantly deteriorates the pseudo-label quality due to an unexpected normalization process. Nevertheless, this phenomenon also could be exploited to facilitate unseen domain generalization. Therefore, to fully exploit the potential of statistics difference while mitigating its negative impacts, we in this paper propose 1) multiple statistics-individual branches to reduce the interference of domain shifts between source domains for reliable pseudo-labels and 2) one statistics-aggregated branch for domain-invariant feature learning. Furthermore, to simulate unseen domains with the statistics difference, we approach this from two aspects, i.e., a perturbation with histogram matching from the image level …

Endoscopic Ultrasound (EUS) is advantageous in perceiving hierarchical changes in the esophageal tract wall for diagnosing submucosal tumors. However, the lesions often disrupt the structural integrity and fine-grained texture information of the esophageal layer, impeding the accurate diagnosis. Moreover, the lesions can appear in any radial position due to the characteristics of EUS imaging, further increasing the difficulty of diagnosis. In this study, we advance an automatic classification model by equipping the Vision Transformer (ViT), a recent state-of-the-art model, with a novel statistical rotation-invariant reinforcement mechanism dubbed SRRM-ViT. Mainly, we adaptively select crucial regions to avoid interference from irrelevant information in the image. Also, this model integrates histogram statistical features with rotation invariance into the self-attention mechanism, achieving bias-free capture of fine-grained information of lesions at arbitrary radial positions. Validated by in-house clinical data and public data, SRRM-ViT has demonstrated remarkable performance improvements, which demonstrates the efficacy and potential of our approach in EUS image classification. Keywords: Fine-Grained Visual Classification (FGVC), Endoscopic Ultrasound (EUS), Rotation Invariant, Token Selection.

Developing self-supervised learning (SSL) models that can learn universal and transferable representations of H&E gigapixel whole-slide images (WSIs) is becoming increasingly valuable in computational pathology. These models hold the potential to advance critical tasks such as few-shot classification, slide retrieval, and patient stratification. Existing approaches for slide representation learning extend the principles of SSL from small images (e.g., 224 x 224 patches) to entire slides, usually by aligning two different augmentations (or \emph{views}) of the slide. Yet the resulting representation remains constrained by the limited clinical and biological relevance of the views. Instead, we postulate that slides stained with multiple markers, such as immunohistochemistry or special stains, can be seen as different views of the same tissue and can constitute a rich task-agnostic training signal. To this end, we introduce MADELEINE, a multimodal pretraining strategy for slide representation learning. MADELEINE is trained with a dual global-local cross-stain alignment objective on large cohorts of breast cancer samples (N=4,211 WSIs across five stains) and kidney transplant samples (N=12,070 WSIs across four stains). We demonstrate the superior quality of slide representations learned by MADELEINE on various downstream evaluations, ranging from morphological and molecular classification to prognostic prediction, on a total of 21 tasks …

Contrastive Language-Image Pre-training (CLIP) has shown its proficiency in acquiring distinctive visual representations and exhibiting strong generalization across diverse vision tasks. However, its effectiveness in pathology image analysis, particularly with limited labeled data, remains an ongoing area of investigation due to challenges associated with significant domain shifts and catastrophic forgetting. Thus, it is imperative to devise efficient adaptation strategies in this domain to enable scalable analysis. In this study, we introduce Path-CLIP, a framework tailored for a swift adaptation of CLIP to various pathology tasks. Firstly, we propose Residual Feature Refinement (RFR) with a dynamically adjustable ratio to effectively integrate and balance source and task-specific knowledge. Secondly, we introduce Hidden Representation Perturbation (HRP) and Dual-view Vision Contrastive (DVC) techniques to mitigate overfitting issues. Finally, we present the Doublet Multimodal Contrastive Loss (DMCL) for fine-tuning CLIP for pathology tasks. We demonstrate that Path-CLIP adeptly adapts pre-trained CLIP to downstream pathology tasks, yielding competitive results. Specifically, Path-CLIP achieves over +19% improvement in accuracy when utilizing mere 0.1% of labeled data in PCam with only 10 minutes of fine-tuning while running on a single GPU.

Radiology reports provide detailed descriptions of medical imaging integrated with patients' medical histories, while report writing is traditionally labor-intensive, increasing radiologists' workload and the risk of diagnostic errors. Recent efforts in automating this process seek to mitigate these issues by enhancing accuracy and clinical efficiency. Emerging research in automating this process promises to alleviate these challenges by reducing errors and streamlining clinical workflows. However, existing automated approaches are based on a single timestamp and often neglect the critical temporal aspect of patients' imaging histories, which is essential for accurate longitudinal analysis. To address this gap, we propose a novel History Enhanced Radiology Report Generation (HERGen) framework that employs a employs a group causal transformer to efficiently integrate longitudinal data across patient visits. Our approach not only allows for comprehensive analysis of varied historical data but also improves the quality of generated reports through an auxiliary contrastive objective that aligns image sequences with their corresponding reports. More importantly, we introduce a curriculum learning-based strategy to adeptly handle the inherent complexity of longitudinal radiology data and thus stabilize the optimization of our framework. The extensive evaluations across three datasets demonstrate that our framework surpasses existing methods in generating accurate radiology reports and …

Defect inspection is paramount within the closed-loop manufacturing system. However, existing datasets for defect inspection often lack the precision and semantic granularity required for practical applications. In this paper, we introduce the Defect Spectrum, a comprehensive benchmark that offers precise, semantic-abundant, and large-scale annotations for a wide range of industrial defects. Building on four key industrial benchmarks, our dataset refines existing annotations and introduces rich semantic details, distinguishing multiple defect types within a single image. With our dataset, we were able to achieve an increase of 10.74% in the Recall rate, and a decrease of 33.10% in the False Positive Rate (FPR) from the industrial simulation experiment. Furthermore, we introduce Defect-Gen, a two-stage diffusion-based generator designed to create high-quality and diverse defective images, even when working with limited defective data. The synthetic images generated by Defect-Gen significantly enhance the performance of defect segmentation models, achieving an improvement in mIoU scores up to 9.85 on Defect-Spectrum subsets. Overall, The Defect Spectrum dataset demonstrates its potential in defect inspection research, offering a solid platform for testing and refining advanced models. Our codes and datasets are released in https://envision-research.github.io/Defect_Spectrum.

Object detection is a pivotal task in computer vision, focused upon localising and categorising objects within the distribution for which they are trained. Nonetheless, the capability of an object detector to localise objects out of the training distribution remains largely unexplored. Whilst recent approaches in object-level out-of-distribution (OoD) detection heavily rely on class-wise labels, such approaches contradict truly open-world scenarios where the number of classes is often unknown. In this context, anomaly detection focuses on detecting unseen instances rather than identifying an object as OoD. This work aims to bridge this gap by leveraging an open-world object detector in conjunction with a self-supervised OoD detector via virtual outlier synthesis. This is achieved by using the detector backbone features to first learn object pseudo-classes in an unsupervised manner. Subsequently, these pseudo-classes serve as the basis for the class-conditional virtual outlier sampling of anomalous features that are classified by an OoD head. Our approach empowers our overall object detector architecture to learn anomaly-aware feature representations without relying on class labels, hence enabling truly open-world object anomaly detection. Empirical validation of our approach demonstrates its effectiveness across diverse datasets encompassing various imaging modalities (visible, infrared, and X-ray). Moreover, our method establishes state-of-the-art performance …

Zero-shot anomaly detection (ZSAD) targets the identification of anomalies within images from arbitrary novel categories. This study introduces AdaCLIP for the ZSAD task, leveraging a pre-trained vision-language model (VLM), CLIP. AdaCLIP incorporates learnable prompts into CLIP and optimizes them through training on auxiliary annotated anomaly detection data. Two types of learnable prompts are proposed: \textit{static} and \textit{dynamic}. Static prompts are shared across all images, serving to preliminarily adapt CLIP for ZSAD. In contrast, dynamic prompts are generated for each test image, providing CLIP with dynamic adaptation capabilities. The combination of static and dynamic prompts is referred to as hybrid prompts, and yields enhanced ZSAD performance. Extensive experiments conducted across 14 real-world anomaly detection datasets from industrial and medical domains indicate that AdaCLIP outperforms other ZSAD methods and can generalize better to different categories and even domains. Finally, our analysis highlights the importance of diverse auxiliary data and optimized prompts for enhanced generalization capacity. Code is available at \texttt{removed for blind review}.

Unified anomaly detection (AD) is one of the most challenges for anomaly detection, where one unified model is trained with normal samples from multiple classes with the objective to detect anomalies in these classes. For such a challenging task, popular normalizing flow (NF) based AD methods may fall into a homogeneous mapping'' issue, where the NF-based AD models are biased to generate similar latent representations for both normal and abnormal features, and thereby lead to a high missing rate of anomalies. In this paper, we propose a novel Hierarchical Gaussian mixture normalizing flow modeling method for accomplishing unified Anomaly Detection, which we call HGAD. Our HGAD consists of two key components: inter-class Gaussian mixture modeling and intra-class mixed class centers learning. Compared to the previous NF-based AD methods, the hierarchical Gaussian mixture modeling approach can bring stronger representation capability to the latent space of normalizing flows, so that even complex multi-class distribution can be well represented and learned in the latent space. In this way, we can avoid mapping different class distributions into the same single Gaussian prior, thus effectively avoiding or mitigating thehomogeneous mapping'' issue. We further indicate that the more distinguishable different class centers, the more conducive …

Recent advancements in end-to-end image compression demonstrate the potential to surpass traditional codecs regarding rate-distortion performance. However, current methods either prioritize human perceptual quality or solely optimize for one or a few predetermined downstream tasks, neglecting a more common scenario that involves a variety of unforeseen machine vision tasks. In this paper, we propose a Diffusion-Based Multiple-Task Unified Image Compression framework that aims to expand the boundary of traditional image compression by incorporating human perception and multiple vision tasks in open-set scenarios. Our proposed method comprises a Multi-task Collaborative Embedding module and a Diffusion-based Invariant Knowledge Learning module. The former module facilitates collaborative embedding for multiple tasks, while the latter module distills the invariant knowledge from seen tasks to generalize toward unseen machine vision tasks. Experiments and visualizations show that the proposed method effectively extracts compact and versatile embeddings for human and machine vision collaborative compression tasks, resulting in superior performance. Specifically, our method outperforms the state-of-the-art by 52.25%/51.68%/48.87%/48.07%/6.29% BD-rate reduction in terms of mAP/mAP/aAcc/PQ-all/accuracy on the MS-COCO for instance detection/instance segmentation/stuff segmentation/panoptic segmentation and video question answering tasks, respectively.

Spiking Neural Networks (SNNs) offer a promising avenue for energy-efficient computing compared with Artificial Neural Networks (ANNs), closely mirroring biological neural processes. However, this potential comes with inherent challenges in directly training SNNs through spatio-temporal backpropagation --- stemming from the temporal dynamics of spiking neurons and their discrete signal processing --- which necessitates alternative ways of training, most notably through ANN-SNN conversion. In this work, we introduce a lightweight Forward Temporal Bias Correction (FTBC) technique, aimed at enhancing conversion accuracy without the computational overhead. We ground our method on provided theoretical findings that through proper temporal bias calibration the expected error of ANN-SNN conversion can be reduced to be zero after each time step. We further propose a heuristic algorithm for finding the temporal bias only in the forward pass, thus eliminating the computational burden of backpropagation and we evaluate our method on CIFAR-10/100 and ImageNet datasets, achieving a notable increase in accuracy on all datasets. Codes are released at a GitHub repository.

Efficient deployment of modern deep convolutional neural networks on resource-constrained devices suffers from demanding computational requirements of convolution operations. Quantization and use of Winograd convolutions operating on sufficiently large-tile inputs are two powerful strategies to speed up convolution operations. However, their combination results in numerical instability, which manifests itself in a strong quality performance degradation. We present an efficient learning scenario that either completely overcomes or strongly reduces the accuracy degradation of full 8-bit quantized F(4, 3) and F(6, 3) Winograd convolutions. Within the global particle swarm optimization (PSO), we derived a set of quantization-friendly Winograd transformations. Following the state-of-the-art (SOTA) training pipeline, we treat Winograd transformations as learnable parameters during network training. Evolving transformations starting from our PSO-derived ones rather than the standard Winograd transformations results in significant numerical error reduction and accuracy improvement. As a consequence, our approach significantly outperforms SOTA methods on various tasks.

Today's deep learning methods focus on how to design the most appropriate objective functions so that the prediction results of the model can be closest to the ground truth. Meanwhile, an appropriate architecture that can facilitate acquisition of enough information for prediction has to be designed. Existing methods ignore a fact that when input data undergoes layer-by-layer feature extraction and spatial transformation, large amount of information will be lost. This paper will delve into the important issues of data loss when data is transmitted through deep networks, namely information bottleneck and reversible functions. We proposed the concept of programmable gradient information (PGI) to cope with the various changes required by deep networks to achieve multiple objectives. PGI can provide complete input information for the target task to calculate objective function, so that reliable gradient information can be obtained to update network weights. In addition, a new lightweight network architecture -- Generalized Efficient Layer Aggregation Network (GELAN), based on gradient path planning is designed. GELAN's architecture confirms that PGI has gained superior results on lightweight models. We verified the proposed GELAN and PGI on MS COCO dataset based object detection. The results show that GELAN only uses conventional convolution operators to …

Structural re-parameterization (SRP) is a novel technique series that boosts neural networks without introducing any computational costs in inference stage. The existing SRP methods have successfully considered many architectures, such as normalizations, convolutions, etc. However, the widely used but computationally expensive attention modules cannot be directly implemented by SRP due to the inherent multiplicative manner and the modules' output is input-dependent during inference. In this paper, we statistically discover a counter-intuitive phenomenon \textit{Stripe Observation} in various settings, which reveals that channel attention values consistently approach some constant vectors during training. It inspires us to propose a novel attention-alike SRP, called ASR, that allows us to achieve SRP for a given network while enjoying the effectiveness of the attention mechanism. Extensive experiments conducted on several standard benchmarks show the effectiveness of ASR in generally improving the performance of various scenarios without any elaborated model crafting. We also provide experimental and theoretical evidence for how the proposed ASR can enhance model performance.

Discriminative models for object classification typically learn image-based representations that do not capture the compositional and 3D nature of objects. In this work, we show that explicitly integrating 3D compositional object representations into deep networks for image classification leads to a largely enhanced generalization in out-of-distribution scenarios. In particular, we introduce a novel architecture, referred to as NOVUM, that consists of a feature extractor and a neural object volume for every target object class. Each neural object volume is a composition of 3D Gaussians that emit feature vectors. This compositional object representation allows for a highly robust and fast estimation of the object class by independently matching the features of the 3D Gaussians of each category to features extracted from an input image. Additionally, the object pose can be estimated via inverse rendering of the corresponding neural object volume. To enable the classification of objects, the neural features at each 3D Gaussian are trained discriminatively to be distinct from (i) the features of 3D Gaussians in other categories, (ii) features of other 3D Gaussians of the same object, and (iii) the background features. Our experiments show that NOVUM offers intriguing advantages over standard architectures due to the 3D compositional structure …

Large-scale self-supervised pre-training has paved the way for one foundation model to handle many different vision tasks. Most pre-training methodologies train a single model of a certain size at one time. Nevertheless, various computation or storage constraints in real-world scenarios require substantial efforts to develop a series of models with different sizes to deploy. Thus, in this study, we propose a novel tri-branch self-supervised training framework, termed as POA (Pre-training Once for All), to tackle this aforementioned issue. Our approach introduces an innovative elastic student branch into a modern self-distillation paradigm. At each pre-training step, we randomly sample a sub-network from the original student to form the elastic student and train all branches in a self-distilling fashion. Once pre-trained, POA allows the extraction of pre-trained models of diverse sizes for downstream tasks. Remarkably, the elastic student facilitates the simultaneous pre-training of multiple models with different sizes, which also acts as an additional ensemble of models of various sizes to enhance representation learning. Extensive experiments, including k-nearest neighbors, linear probing evaluation and assessments on downstream tasks demonstrate the effectiveness and advantages of our POA. It achieves state-of-the-art performance using ViT, Swin Transformer and ResNet backbones, producing around a hundred models with …

Multi-exit network is a promising architecture for efficient model inference by sharing backbone networks and weights among multiple exits. However, the gradient conflict of the shared weights results in sub-optimal accuracy. This paper introduces Deep Feature Surgery (DFS), which consists of feature partitioning and feature referencing approaches to resolve gradient conflict issues during the training of multi-exit networks. The feature partitioning separates shared features along the depth axis among all exits to alleviate gradient conflict while simultaneously promoting joint optimization for each exit. Subsequently, feature referencing enhances multi-scale features for distinct exits across varying depths to improve the model accuracy. Furthermore, DFS reduces the training operations with the reduced complexity of backpropagation. Experimental results on Cifar100 and ImageNet datasets exhibit that DFS provides up to a 50.00% reduction in training time and attains up to a 6.94% enhancement in accuracy when contrasted with baseline methods across diverse models and tasks. Budgeted batch classification evaluation on MSDNet demonstrates that DFS uses about 2x fewer average FLOPs per image to achieve the same classification accuracy as baseline methods on Cifar100.

Domain generalization (DG) aims to avoid the performance degradation of the model when the distribution shift between the limited training data and unseen test data occurs. Recently, foundation models with enormous parameters have been pre-trained with huge datasets, demonstrating strong generalization ability and showing promising direction for solving the DG problem. However, fully Fine-Tuning (FT) the foundation models results in unsatisfactory out-of-distribution accuracy due to the destroyed pre-trained generalized features. Recently, Parameter-Efficient Fine-Tuning (PEFT) alleviates the above problem by fine-tuning a small portion of the model parameters while keeping the rest frozen, which achieves better generalization performance compared to FT. Nevertheless, PEFT still suffers from the issue of overfitting to the training domains. To address the above issue, we propose Parameter-Efficient Group with Orthogonal regularization (PEGO) for vision transformers, which effectively preserves the generalization ability of the pre-trained network and learns more diverse knowledge compared with conventional PEFT. Specifically, we inject a group of trainable Low-Rank Adaptation (LoRA) modules into the pre-trained model and propose an orthogonal regularization loss to enhance the generalization ability of the model. Our framework achieves SOTA performance on five DG benchmarks, while only requiring training a small number of parameters without adding additional testing cost.

Federated learning presents massive potential for privacy-friendly vision task collaboration. However, the federated visual performance is deeply affected by backdoor attacks, where malicious clients optimize on triggered samples to mislead the global model into targeted mispredictions. Existing backdoor defensive solutions are normally based on two assumptions: data homogeneity and minority malicious ratio for the elaborate client-wise defensive rules. To address existing limitations, we argue that heterogeneous clients and backdoor attackers both bring divergent optimization directions and thus it is hard to discriminate them precisely. In this paper, we argue that parameters appear in different important degrees towards distinct distribution and instead consider meaningful and meaningless parameters for the ideal target distribution. We propose the Self-Driven Fisher Calibration (SDFC), which utilizes the Fisher Information to calculate the parameter importance degree for the local agnostic and global validation distribution and regulate those elements with large important differences. Furthermore, we allocate high aggregation weight for clients with relatively small overall parameter differences, which encourages clients with close local distribution to the global distribution, to contribute more to the federation. This endows SDFC to handle backdoor attackers in heterogeneous federated learning. Various vision task performances demonstrate the effectiveness of SDFC.

Machine Unlearning removes specific knowledge about training data samples corresponding effects from an already trained model. It has significant practical benefits, such as purging private, inaccurate, or outdated information from trained models without the need for complete re-training. Unlearning within a multimodal setting presents unique challenges due to the intrinsic ependencies between different data modalities and the expensive cost of training on large multimodal datasets and architectures. This paper presents the first machine unlearning approach for multimodal data and models, titled MultiDelete, which is designed to decouple associations between unimodal data points during unlearning without losing the overall representation strength of the trained model. MultiDelete advocates for three key properties for effective multimodal unlearning: (a): modality decoupling, which effectively decouples the association between individual unimodal data points marked for deletion, rendering them as unrelated data points, (b): multimodal knowledge retention, which retains the multimodal representation capability of the model post-unlearning, and (c): unimodal knowledge retention, which retains the unimodal representation capability of the model post-unlearning. MultiDelete is efficient to train and is not constrained by using a strongly convex loss--a common restriction among many existing baselines. Experiments on two multimodal architectures and four datasets, including image-text and graph-text datasets, show …

Self-supervised learning has recently emerged as the preeminent pretraining paradigm across and between modalities, with remarkable results. In the image domain specifically, group (or cluster) discrimination has been one of the most successful methods. However, such frameworks need to guard against heavily imbalanced cluster assignments to prevent collapse to trivial solutions. Existing works typically solve this by reweighing cluster assignments to promote balance, or with offline operations (e.g. regular re-clustering) that prevent collapse. However, the former typically requires large batch sizes, which leads to increased resource requirements, and the latter introduces scalability issues with regard to large datasets. In this work, we propose ExCB, a framework that tackles this problem with a novel cluster balancing method. ExCB estimates the relative size of the clusters across batches and balances them by adjusting cluster assignments, proportionately to their relative size and in an online manner. Thereby, it overcomes previous methods' dependence on large batch sizes and is fully online, and therefore scalable to any dataset. We conduct extensive experiments to evaluate our approach and demonstrate that ExCB: a) achieves state of the art results with significantly reduced resource requirements compared to previous works, b) is fully online, and therefore scalable to large …

Contrastive Language Image Pre-training (CLIP) has recently demonstrated success across various tasks due to superior feature representation empowered by image-text contrastive learning. However, the instance discrimination method used by CLIP can hardly encode the semantic structure of training data. To handle this limitation, cluster discrimination has been proposed through iterative cluster assignment and classification. Nevertheless, most cluster discrimination approaches only define a single pseudo-label for each image, neglecting multi-label signals in the image. In this paper, we propose a novel Multi-Label Cluster Discrimination method named MLCD to enhance representation learning. In the clustering step, we first cluster the large-scale LAION-400M dataset into one million centers based on off-the-shelf embedding features. Considering that natural images frequently contain multiple visual objects or attributes, we select the multiple closest centers as auxiliary class labels. In the discrimination step, we design a novel multi-label classification loss, which elegantly separates losses from positive classes and negative classes, and alleviates ambiguity on decision boundary. We validate the proposed multi-label cluster discrimination method with experiments on different scales of models and pre-training datasets. Experimental results show that our method achieves state-of-the-art performance on multiple downstream tasks including linear probe, zero-shot classification, and image-text retrieval.

Robust fine-tuning aims to adapt a vision-language model to downstream tasks while preserving its zero-shot capabilities on unseen data. Recent studies have introduced fine-tuning strategies to improve in-distribution (ID) performance on the downstream tasks while minimizing deterioration in out-of-distribution (OOD) performance on unseen data. This balance is achieved either by aligning the fine-tuned representations with the pre-trained ones or by constraining significant deviations in fine-tuned weights compared to the pre-trained model. In the latter approach, the regularization term is uniformly applied to all parameters. Our work proposes to selectively apply the regularization term based on the ``importance'' of each neuron to the fine-tuning dataset. To this end, we develop an importance-score metric to quantify each neurons’ importance to the downstream task and then leverage this to develop two fine-tuning strategies: importance-guided selective fine-tuning and importance-guided regularization. Our approach can be used concurrently with representation space-based methods, outperforming other approaches based on parameter space. We improve the state-of-the-art on standard robust fine-tuning benchmarks across datasets in both the full-shot and low-shot settings.

The widespread use of pre-trained large language models (and multi-modal models) has led to powerful performance across a wide range of tasks. Despite their effectiveness, we have limited knowledge of their internal knowledge representation. Motivated by and using the classic problem of Compositional Zero-Shot Learning (CZSL) as an example, we provide a structured view of the latent space that any general model (LLM or otherwise) should nominally respect. Based on this view, we first provide a practical solution to the CZSL problem based on a cross-attention mechanism that can deal with both Open and Closed-World single-attribute compositions as well as multi-attribute compositions with relative ease. In all three tasks, our approach yields performance competitive with methods designed solely for that task (i.e., adaptations to other tasks are difficult). Then, we extend this perspective to existing LLMs and ask to what extent they satisfy our axiomatic definitions. Our analysis shows a mix of interesting and unsurprising findings, but nonetheless suggests that our criteria is meaningful and may yield a more structured approach to training such models, strategies for additional data collection, and diagnostics beyond visual inspection.

We combine concept-based neural networks with generative, flow-based classifiers into a novel, intrinsically explainable, exactly invertible approach to supervised learning. Prototypical neural networks, a type of concept-based neural network, represent an exciting way forward in realizing human-comprehensible machine learning without concept annotations, but a human-machine semantic gap continues to haunt current approaches. We find that reliance on indirect interpretation functions for prototypical explanations imposes a severe limit on prototypes' informative power. From this, we posit that invertibly learning prototypes as distributions over the latent space provides more robust, expressive, and interpretable modeling. We propose one such model, called ProtoFlow, by composing a normalizing flow with Gaussian mixture models. ProtoFlow (1) sets a new state-of-the-art in joint generative and predictive modeling and (2) achieves predictive performance comparable to existing prototypical neural networks while enabling richer interpretation.

In the realm of current unsupervised image outlier detection, assigning outlier scores holds greater significance than its subsequent task: thresholding for predicting labels. This is because determining the optimal threshold on non-separable outlier score functions is an ill-posed problem. However, the lack of predicted labels not only hinders some real applications of current outlier detectors but also causes these methods not able to be enhanced by leveraging the dataset’s self-supervision. To advance existing scoring methods, we propose a multiple thresholding (Multi-T) module. It generates two thresholds that isolate inliers and outliers from the unlabelled target dataset, whereas outliers are employed to obtain better feature representation while inliers provide an uncontaminated manifold. Extensive experiments verify that Multi-T can significantly improve proposed outlier scoring methods. Moreover, Multi-T contributes to a naive distance-based method being state-of-the-art.

The inability of deep learning models to handle data drawn from unseen distributions has sparked much interest in unsupervised out-of-distribution (U-OOD) detection, as it is crucial for reliable deep learning models. Despite considerable attention, theoretically-motivated approaches are few and far between, with most methods building on top of some form of heuristic. Recently, U-OOD was formalized in the context of data invariants, allowing a clearer understanding of how to characterize U-OOD, and methods leveraging affine invariants have attained state-of-the-art results on large-scale benchmarks. Nevertheless, the restriction to affine invariants hinders the expressiveness of the approach. In this work, we broaden the affine invariants formulation to a more general case and propose a framework consisting of a normalizing flow-like architecture capable of learning non-linear invariants. Our novel approach achieves state-of-the-art results on an extensive U-OOD benchmark, and we demonstrate its further applicability to tabular data. Finally, we show our method has the same desirable properties as those based on affine invariants.

Conventional multi-label recognition methods often focus on label confidence, frequently overlooking the pivotal role of partial order relations consistent with human preference. To resolve these issues, we introduce a novel method for multimodal label relevance ranking, named Label Relevance Ranking with Proximal Policy Optimization (LR\textsuperscript{2}PPO), which effectively discerns partial order relations among labels. LR\textsuperscript{2}PPO first utilizes partial order pairs in the target domain to train a reward model, which aims to capture human preference intrinsic to the specific scenario. Furthermore, we meticulously design state representation and a policy loss tailored for ranking tasks, enabling LR\textsuperscript{2}PPO to boost the performance of label relevance ranking model and largely reduce the requirement of partial order annotation for transferring to new scenes. To assist in the evaluation of our approach and similar methods, we further propose a novel benchmark dataset, LRMovieNet, featuring multimodal labels and their corresponding partial order data. Extensive experiments demonstrate that our LR\textsuperscript{2}PPO algorithm achieves state-of-the-art performance, proving its effectiveness in addressing the multimodal label relevance ranking problem. Codes and the proposed LRMovieNet dataset are publicly available at \url{https://github.com/ChazzyGordon/LR2PPO}.

The partial label challenge in Multi-Label Class-Incremental Learning (MLCIL) arises when only the new classes are labeled during training, while past and future labels remain unavailable. This issue leads to a proliferation of false-positive errors due to erroneously high confidence multi-label predictions, exacerbating catastrophic forgetting within the disjoint label space. In this paper, we aim to refine multi-label confidence calibration in MLCIL and propose a Confidence Self-Calibration (CSC) approach. Firstly, for label relationship calibration, we introduce a class-incremental graph convolutional network that bridges the isolated label spaces by constructing learnable, dynamically extended label relationship graph. Then, for confidence calibration, we present a max-entropy regularization for each multi-label increment, facilitating confidence self-calibration through the penalization of over-confident output distributions. Our approach attains new state-of-the-art results in MLCIL tasks on both MS-COCO and PASCAL VOC datasets, with the calibration of label confidences confirmed through our methodology.

Learning from noisy data is a challenging task, as noisy labels can compromise decision boundaries and result in suboptimal generalization performance. Most previous approaches for dealing noisy data are based on sample selection, which utilized the small loss criterion to reduce the adverse effects of noisy labels. Nevertheless, they encounter a critical limitation in being unable to effectively separate challenging samples from those that were merely mislabeled. To this end, we propose a novel moving trace and feature density-based confidence sample selection strategy (called MTaDCS). Different from existing small loss-based approaches, the local feature density of samples in the latent space is explored to construct a confidence set by selectively choosing confident samples in a progressive manner in terms of moving trace. Therefore, our MTaDCS can gradually isolate noisy labels through the setting of confidence set and achieve the goal of learning discriminative features from hard samples. Extensive experiments conducted on datasets with simulated and real-world noises validate that the proposed MTaDCS outperforms the state-of-the-art methods in terms of various metrics. We will make our code publicly available.

Active learning (AL) in open set scenarios presents a novel challenge of identifying the most valuable examples in an unlabeled data pool that comprises data from both known and unknown classes. Traditional methods prioritize selecting informative examples with low confidence, with the risk of mistakenly selecting unknown-class examples with similarly low confidence. Recent methods favor the most probable known-class examples, with the risk of picking simple already mastered examples. In this paper, we attempt to query examples that are both likely from known classes and highly informative, and propose a \textit{Bidirectional Uncertainty-based Active Learning} (BUAL) framework. Specifically, we achieve this by first pushing the unknown class examples toward regions with high-confidence predictions with our proposed \textit{Random Label Negative Learning} method. Then, we propose a \textit{Bidirectional Uncertainty sampling} strategy by jointly estimating uncertainty posed by both positive and negative learning to perform consistent and stable sampling. BUAL successfully extends existing uncertainty-based AL methods to complex open-set scenarios. Extensive experiments on multiple datasets with varying openness demonstrate that BUAL achieves state-of-the-art performance.

With the advancement of deep neural networks in computer vision, artificial intelligence (AI) is widely employed in real-world applications. However, AI still faces limitations in mimicking high-level human capabilities, such as novel category discovery, for practical use. While some methods utilizing offline continual learning have been proposed for novel category discovery, they neglect the continuity of data streams in real-world settings. In this work, we introduce Online Continuous Generalized Category Discovery (OCGCD), which considers the dynamic nature of data streams where data can be created and deleted in real time. Additionally, we propose a novel method, DEAN, Discovery via Energy guidance and feature AugmentatioN, which can discover novel categories in an online manner through energy-guided discovery and facilitate discriminative learning via energy-based contrastive loss. Furthermore, DEAN effectively pseudo-labels unlabeled data through variational feature augmentation. Experimental results demonstrate that our proposed DEAN achieves outstanding performance in proposed OCGCD scenario. The code will be publicly released after the review.

Open-set domain adaptation aims to improve the generalization performance of a learning algorithm on a more realistic problem of open-set domain shift where the target data contains an additional unknown class that is not present in the source data. Most existing algorithms include two phases that can be described as closed-set domain adaptation given heuristic unknown class separation. Therefore, the generalization error cannot be strictly bounded due to the gap between the true distribution and samples inferred from heuristics. In this paper, we propose an end-to-end algorithm that tightly bound the risk of the entire target task by positive-unlabeled (PU) learning theory and the joint error from domain adaptation. Extensive experiments on various data sets demonstrate the effectiveness and efficiency of our proposed algorithm over open-set domain adaptation baselines.

In this work, we delve deep into the diverse factors that influence the efficacy of modern unsupervised domain adaptation (UDA) methods using a large-scale, controlled empirical study. To facilitate our analysis, we first develop UDA-Bench, a novel PyTorch framework that standardizes training and evaluation for domain adaptation enabling fair comparisons across several UDA methods. Using UDA-Bench, our comprehensive empirical study into the impact of backbone architectures, unlabeled data quantity, and pre-training datasets reveals that: (i) the benefits of adaptation methods diminish with advanced backbones, (ii) current methods underutilize unlabeled data, and (iii) pre-training data significantly affects downstream adaptation in both supervised and self-supervised settings. In the context of unsupervised adaptation, these observations uncover several novel and surprising properties, while scientifically validating several others that were often considered empirical heuristics or practitioner intuitions in the absence of a standardized training and evaluation framework. The UDA-Bench framework and trained models will be publicly released to promote future UDA research.

Few-shot class-incremental learning (FSCIL) aims to learn sequential classes with limited samples in a few-shot fashion. Inherited from the classical class-incremental learning setting, the popular benchmark of FSCIL uses averaged accuracy (aAcc) and last-task averaged accuracy (lAcc) as the evaluation metrics. However, we reveal that such evaluation metrics may not provide adequate emphasis on the novel class performance, and the continual learning ability of FSCIL methods could be ignored under this benchmark. In this work, as a complement to existing metrics, we offer a new metric called generalized average accuracy (gAcc) which is designed to provide an extra equitable evaluation by incorporating different perspectives of the performance under the guidance of a parameter α. We also present an overall metric in the form of the area under the curve (AUC) along the α. Under the guidance of gAcc, we release the potential of intermediate features of the vision transformers to boost the novel-class performance. Taking information from intermediate layers which are less class-specific and more generalizable, we manage to rectify the final features, leading to a more generalizable transformer-based FSCIL framework. Without complex network designs or cumbersome training procedures, our method outperforms existing FSCIL methods at aAcc and gAcc on …

Incremental Learning (IL) aims to accumulate knowledge from sequential input tasks while overcoming catastrophic forgetting. Existing IL methods typically assume that an incoming task has only increments of classes or domains, referred to as Class IL (CIL) or Domain IL (DIL), respectively. In this work, we consider a more challenging and realistic but under-explored IL scenario, named Versatile Incremental Learning (VIL), in which a model has no prior of which of the classes or domains will increase in the next task. In the proposed VIL scenario, the model faces intra-class domain confusion and inter-domain class confusion, which make the model fail to accumulate new knowledge without interference with learned knowledge. To address these issues, we propose a simple yet effective IL framework, named Incremental Classifier with Adaptation Shift cONtrol (ICON). Based on shifts of learnable modules, we design a novel regularization method called Cluster-based Adaptation Shift conTrol (CAST) to control the model to avoid confusion with the previously learned knowledge and thereby accumulate the new knowledge more effectively. Moreover, we introduce Incremental Classifier (IC) which expands its output nodes to address the overwriting issue from different domains corresponding to a single class while maintaining the previous knowledge. We conducted extensive …

Prompt-tuning methods for Continual Learning (CL) freeze a large pre-trained model and focus training on a few parameter vectors termed prompts. Most of these methods organize these vectors in a pool of key-value pairs, and use the input image as query to retrieve the prompts (values). However, as keys are learned while tasks progress, the prompting selection strategy is itself subject to catastrophic forgetting, an issue often overlooked by existing approaches. For instance, prompts introduced to accommodate new tasks might end up interfering with previously learned prompts. To make the selection strategy more stable, we ask a foundational model (CLIP) to select our prompt within a two-level adaptation mechanism. Specifically, the first level leverages standard textual prompts for the CLIP textual encoder, leading to stable class prototypes. The second level, instead, uses these prototypes along with the query image as keys to index a second pool. The retrieved prompts serve to adapt a pre-trained ViT, granting plasticity. In doing so, we also propose a novel residual mechanism to transfer CLIP semantics to the ViT layers. Through extensive analysis on established CL benchmarks, we show that our method significantly outperforms both state-of-the-art CL approaches and the zero-shot CLIP test. Notably, our …

In this paper, we propose a new knowledge transfer paradigm called Knowledge in One Prompt (KiOP). This paradigm encapsulates knowledge from various models into a solitary prompt without altering the original models or requiring access to the training data, which enables us to achieve efficient and convenient knowledge transfer in more realistic scenarios. From a practicality standpoint, this paradigm not only for the first time proves the effectiveness of Visual Prompt in data inaccessible contexts, but also solves the problems of low model reusability and high storage resource consumption faced by traditional Data-Free Knowledge Transfer, which means that we can realize the parallel knowledge transfer of multiple models without modifying any source model. Extensive experiments across various datasets and models demonstrate the efficacy of the proposed KiOP knowledge transfer paradigm. Without access to real training data and with rigorous storage capacity constraints, it is also capable of yielding considerable outcomes when dealing with cross-model backbone setups and handling parallel knowledge transfer processing requests with multiple (more than 2) models.

Domain Generalization (DG) focuses on enhancing the generalization of deep learning models trained on multiple source domains to adapt to unseen target domains. This paper explores DG through the lens of bias-variance decomposition, uncovering that test errors in DG predominantly arise from cross-domain bias and variance. Inspired by this insight, we introduce a Representation Enhancement-Stabilization (RES) framework, comprising a Representation Enhancement (RE) module and a Representation Stabilization (RS) module. In RE, a novel set of feature frequency augmentation techniques is used to progressively reduce cross-domain bias during feature extraction. Furthermore, in RS, a novel Mutual Exponential Moving Average (MEMA) strategy is designed to stabilize model optimization for diminishing cross-domain variance during training. Collectively, the whole RES method can significantly enhance model generalization. We evaluate RES on five benchmark datasets and the results show that it outperforms multiple advanced DG methods. Our code will be publicly available.

Knowledge Distillation (KD) has proven effective for compressing large teacher models into smaller student models. While it is well known that student models can achieve similar accuracies as the teachers, it has also been shown that they nonetheless often do not learn the same function. It is, however, often highly desirable that the student’s and teacher’s functions share similar properties such as basing the prediction on the same input features, as this ensures that students learn the ‘right features’ from the teachers. In this work, we explore whether this can be achieved by not only optimizing the classic KD loss but also the similarity of the explanations generated by the teacher and the student. Despite the idea being simple and intuitive, we find that our proposed ‘explanation-enhanced’ KD (e2KD) (1) consistently provides large gains in terms of accuracy and student-teacher agreement, (2) ensures that the student learns from the teacher to be right for the right reasons and to give similar explanations, and (3) is robust with respect to the model architectures, the amount of training data, and even works with ‘approximate’, pre-computed explanations.

Recently, the scale of transformers has grown rapidly, which introduces considerable challenges in terms of training overhead and inference efficiency in the scope of task adaptation. Existing works, namely Parameter-Efficient Fine-Tuning (PEFT) and model compression, have separately investigated the challenges. However, PEFT cannot guarantee the inference efficiency of the original backbone, especially for large-scale models. Model compression requires significant training costs for structure searching and re-training. Consequently, a simple combination of them cannot guarantee accomplishing both training efficiency and inference efficiency with minimal costs. In this paper, we propose a novel Parallel Yielding Re-Activation (PYRA) method for such a challenge of training-inference efficient task adaptation. PYRA first utilizes parallel yielding adaptive weights to comprehensively perceive the data distribution in downstream tasks. A re-activation strategy for token modulation is then applied for tokens to be merged, leading to calibrated token features. Extensive experiments demonstrate that PYRA outperforms all competing methods under both low compression rate and high compression rate, demonstrating its effectiveness and superiority in maintaining both training efficiency and inference efficiency for large-scale foundation models. Our code will be released to the public.

Data-efficient learning has garnered significant attention, especially given the current trend of large multi-modal models. Recently, dataset distillation has become an effective approach for data-efficiency by synthesizing data samples that are most essential for network training; however, it remains to be explored which samples are essential for the dataset distillation process itself. In this work, we study the data efficiency and selection for the dataset distillation task. By re-formulating the dynamics of distillation, we provide insight into the inherent redundancy in the real dataset, both theoretically and empirically. Thus we propose to use the empirical loss value as a static data pruning criterion. To further compensate for the variation of the data value in training, we propose to find the most contributing samples based on their causal effects on the distillation. The proposed selection strategy can efficiently exploit the training dataset and outperform the previous SOTA distillation algorithms, and consistently enhance the distillation algorithms, even on much larger-scale and more heterogeneous datasets, e.g. full ImageNet-1K and Kinetics-400. We believe this paradigm will open up new avenues in the dynamics of distillation and pave the way for efficient dataset distillation. Our code will be made publicly available.

This paper introduces \textbf{RDA}, a pioneering approach designed to address two primary deficiencies prevalent in previous endeavors aiming at stealing pre-trained encoders: (1) suboptimal performances attributed to biased optimization objectives, and (2) elevated query costs stemming from the end-to-end paradigm that necessitates querying the target encoder every epoch. Specifically, we initially \textbf{\underline{R}}efine the representations of the target encoder for each training sample, thereby establishing a less biased optimization objective before the steal-training phase. This is accomplished via a sample-wise prototype, which consolidates the target encoder's representations for a given sample's various perspectives. Demanding exponentially fewer queries compared to the end-to-end approach, prototypes can be instantiated to guide subsequent query-free training. For more potent efficacy, we develop a multi-relational extraction loss that trains the surrogate encoder to \textbf{\underline{D}}iscriminate mismatched embedding-prototype pairs while \textbf{\underline{A}}ligning those matched ones in terms of both amplitude and angle. In this way, the trained surrogate encoder achieves state-of-the-art results across the board in various downstream datasets with limited queries. Moreover, RDA is shown to be robust to multiple widely-used defenses. Our code is available at https://anonymous.4open.science/r/RDA.

Graph neural network (GNN) explainers for graph classification aim to identify the important subgraph that ensures the prediction for a given graph. Until now, most of the existing GNN explainers are based on association, and a few are causality-inspired but they are associated-based in essence. Associated-based explainers are shown to be prone to spurious correlations. We propose CXGNN, a GNN causal explainer via causal inference. Our explainer is based on the observation that a graph often consists of a causal subgraph. Specifically, CXGNN includes three main steps: 1) Building causal structure and the corresponding structural causal model (SCM) for a graph, which enables the cause-effect calculation among nodes. 2) Directly calculating the cause-effect in real-world graphs is computationally challenging. We are then enlightened by the recently proposed neural causal model (NCM), a special type of SCM that is trainable, and the design of customized NCMs for GNNs. By training these GNN NCMs, the cause-effect can be easily calculated. 3) We uncover the subgraph that causally explains the GNN predictions via the well-trained GNN NCMs. Evaluation results on multiple synthetic and real-world graphs validate that CXGNN significantly outperforms the existing GNN explainers in exactly finding the ground-truth explanations.

Uncertainty Attribution (UA) aims to understand the sources of uncertainty in deep learning models by identifying key contributors to predictive uncertainty. To improve the faithfulness of existing UA methods, we formulate UA as an optimization problem to learn a binary mask on the input, identifying regions that significantly contribute to output uncertainty. The learned mask allows uncertainty reduction through learning informative perturbations on the masked input. Our method enhances UA interpretability and maintains high efficiency by integrating three key improvements: Segment Anything Model (SAM)-guided mask parameterization for efficient and interpretable mask learning; learnable perturbations that surpass traditional techniques by adaptively targeting and refining problematic regions specific to each input without the need for manually tuning the perturbation parameters; and a novel application of Gumbel-sigmoid reparameterization for efficiently learning Bernoulli-distributed binary masks under continuous optimization. Our experiments on the detection of problematic regions and faithfulness tests demonstrate our method's superiority over state-of-the-art UA methods.

Existing automated symbolic optimizer design methods necessitate the use of proxies, often resulting in significant performance degradation when transferring to a target domain. In this paper, we propose a learning based model called Symbolic Optimizer Learner (SOL) that can discover high-performance symbolic optimizers directly on the target. SOL is integrated with symbols and can be directly transformed into a symbolic optimizer. In addition, an unrolled optimization approach is introduced for SOL training. SOL can be embedded into the training process of neural networks, optimizing the target directly without any proxies. Our extensive experiments demonstrate the good performance and high generalizability of SOL through diverse tasks, ranging from classifications to adversarial attacks, from GNN to NLP tasks. On image classification, SOL achieved ~5x speedup and ~3\% accuracy gain. On adversarial attacks, SOL achieved the best attack success rate across seven SOTA defense models. On GNN training, SOL discovered optimizers can outperform Adam on three different datasets. On BERT fine-tuning, SOL also outperformed AdamW on five benchmarks.

Generative Adversarial Networks(GANs) have received considerable attention due to its outstanding ability to generate images. However, training a GAN is hard since the game between the Generator(G) and the Discriminator(D) is unfair. Towards making the competition fairer, we propose a new perspective of training GANs, named Consistent Latent Representation and Reconstruction(CLR-GAN). In this paradigm, we treat the G and D as an inverse process, the discriminator has an additional task to restore the pre-defined latent code while the generator also needs to reconstruct the real input, thus obtaining a relationship between the latent space of G and the out-features of D. Based on this prior, we can put D and G on an equal position during training using a new criterion. Experimental results on various datasets and architectures prove our paradigm can make GANs more stable and generate better quality images(31.22% gain of FID on CIFAR10 and 39.5% on AFHQ-Cat}, respectively). We hope that the proposed perspective can inspire researchers to explore different ways of viewing GANs training, rather than being limited to a two-player game. The code will be publicly available soon at [Removed for blind review].

In this paper, we address the challenge of compressing generative adversarial networks (GANs) for deployment in resource-constrained environments by proposing two novel methodologies: Distribution Matching for Efficient compression (DiME) and Network Interactive Compression via Knowledge Exchange and Learning (NICKEL). DiME employs foundation models as embedding kernels for efficient distribution matching, leveraging maximum mean discrepancy to facilitate effective knowledge distillation. Simultaneously, NICKEL employs an interactive compression method that enhances the communication between the student generator and discriminator, achieving a balanced and stable compression process. Our comprehensive evaluation on the StyleGAN2 architecture with the FFHQ dataset shows the effectiveness of our approach, with NICKEL & DiME achieving FID scores of 10.45 and 15.93 at compression rates of 95.73% and 98.92%, respectively. Remarkably, our methods sustain generative quality even at an extreme compression rate of 99.69%, surpassing the previous state-of-the-art performance by a large margin. These findings not only demonstrate our methodologies' capacity to significantly lower GANs' computational demands but also pave the way for deploying high-quality GAN models in settings with limited resources. Our code will be released soon.

Availability poisons exploit supervised learning (SL) algorithms by introducing class-related shortcut features in images such that models trained on poisoned data are useless for real-world datasets. Self-supervised learning (SSL), which utilizes augmentations to learn instance discrimination, is regarded as a strong defense against poisoned data. However, by extending the study of SSL across multiple poisons on the CIFAR-10 and ImageNet-100 datasets, we demonstrate that it often performs poorly, far below that of training on clean data. Leveraging the vulnerability of SL to poison attacks, we introduce adversarial training (AT) on SL to obfuscate poison features and guide robust feature learning for SSL. Our proposed defense, designated VESPR (Vulnerability Exploitation of Supervised Poisoning for Robust SSL), surpasses the performance of six previous defenses across seven popular availability poisons. VESPR displays superior performance over all previous defenses, boosting the minimum and average ImageNet-100 test accuracies of poisoned models by 16% and 9%, respectively. Through analysis and ablation studies, we elucidate the mechanisms by which VESPR learns robust class features. Code for VESPR will be made available upon conference acceptance of this paper.

Does our behavior change when we speak over video calls? Given two unpaired sets of videos of people, we seek to automatically find spatio-temporal patterns that are distinctive of each set. Existing methods use discriminative approaches and perform post-hoc explainability analysis. Such methods are insufficient as they are unable to provide insights beyond obvious dataset biases, and the explanations are useful only if humans themselves are good at the task. Instead, we tackle the problem through the lens of generative domain translation: our method generates a detailed report of learned, input-dependent spatio-temporal features and the extent to which they vary between the domains. We demonstrate that our method can discover behavioral differences between conversing face-to-face (F2F) and on video-calls (VCs). We also show the applicability of our method on discovering differences in presidential communication styles. Additionally, we are able to predict temporal change-points in videos that decouple expressions in an unsupervised way, and increase the interpretability and usefulness of our model. Finally, our method, being generative, can be used to transform a video call to appear as if it were recorded in a F2F setting. Experiments and visualizations show our approach is able to discover a range of behaviors, taking …

Generating medical reports for X-ray images is a challenging task, particularly in an unpaired scenario where paired image-report data is unavailable for training. To address this challenge, we propose a novel model that leverages the available information in two distinct datasets, one comprising reports and the other consisting of images. The core idea of our model revolves around the notion that combining auto-encoding report generation with multi-modal (report-image) alignment can offer a solution. However, the challenge persists regarding how to achieve this alignment when pair correspondence is absent. Our proposed solution involves the use of auxiliary tasks, particularly contrastive learning and classification, to position related images and reports in close proximity to each other. This approach differs from previous methods that rely on pre-processing steps using external information stored in a knowledge graph. Our model, named MedRAT, surpasses previous state-of-the-art methods, demonstrating the feasibility of generating comprehensive medical reports without the need for paired data or external tools.

Achieving disentangled control over multiple facial motions and accommodating diverse input modalities greatly enhances the application and entertainment of the talking head generation. This necessitates a deep exploration of the decoupling space for facial features, ensuring that they a) operate independently without mutual interference and b) can be preserved to share with different modal inputs—both aspects often neglected in existing methods. To address this gap, this paper proposes a novel Efficient Disentanglement framework for Talking head generation (EDTalk). Our framework enables individual manipulation of mouth shape, head pose, and emotional expression, conditioned on video or audio inputs. Specifically, we employ three lightweight modules to decompose the facial dynamics into three distinct latent spaces representing mouth, pose, and expression, respectively. Each space is characterized by a set of learnable bases whose linear combinations define specific motions. To ensure independence and accelerate training, we enforce orthogonality among bases and devise an efficient training strategy to allocate motion responsibilities to each space without relying on external knowledge. The learned bases are then stored in corresponding banks, enabling shared visual priors with audio input. Furthermore, considering the properties of each space, we propose an Audio-to-Motion module for audio-driven talking head synthesis. Experiments are conducted …

Texturing 3D humans with semantic UV maps remains a challenge due to the difficulty of acquiring reasonably unfolded UV. Despite recent text-to-3D advancements in supervising multi-view renderings using large text-to-image (T2I) models, issues persist with generation speed, text consistency, and texture quality, resulting in data scarcity among existing datasets. We present TexDreamer, the first zero-shot multimodal high-fidelity 3D human texture generation model. Utilizing an efficient texture adaptation finetuning strategy, we adapt large T2I model to a semantic UV structure while preserving its original generalization capability. Leveraging a novel feature translator module, the trained model is capable of generating high-fidelity 3D human textures from either text or image within seconds. Furthermore, we introduce ArTicuLated humAn textureS (ATLAS), the largest high-resolution (1, 024 × 1, 024) 3D human texture dataset which contains 50k high-fidelity textures with text descriptions. Our dataset and model will be available for research purposes.

3D content creation has achieved significant progress in terms of both quality and speed. Although current feed-forward models can produce 3D objects in seconds, their resolution is constrained by the intensive computation required during training. In this paper, we introduce Large Multi-View Gaussian Model (LGM), a novel framework designed to generate high-resolution 3D models from text prompts or single-view images. Our key insights are two-fold: 1) 3D Representation: We propose multi-view Gaussian features as an efficient yet powerful representation, which can then be fused together for differentiable rendering. 2) 3D Backbone: We present an asymmetric U-Net as a high-throughput backbone operating on multi-view images, which can be produced from text or single-view image input by leveraging multi-view diffusion models. Extensive experiments demonstrate the high fidelity and efficiency of our approach. Notably, we maintain the fast speed to generate 3D objects within 5 seconds while boosting the training resolution to 512, thereby achieving high-resolution 3D content generation.

Manually creating textures for 3D meshes is time-consuming, even for expert visual content creators. We propose a fast approach for automatically texturing an input 3D mesh based on a user-provided text prompt. Importantly, our approach disentangles lighting from surface material/reflectance in the resulting texture so that the mesh can be properly relit and rendered in any lighting environment. Our method introduces LightControlNet, a new text-to-image model based on the ControlNet architecture, that allows the specification of the desired lighting as a conditioning image to the model. Our text-to-texture pipeline then constructs the texture in two stages. The first stage produces a sparse set of visually consistent reference views of the mesh using LightControlNet. The second stage applies a texture optimization based on Score Distillation Sampling (SDS) that works with LightControlNet to increase the texture quality while disentangling surface material from lighting. We show that this pipeline is significantly faster than previous text-to-texture methods, while producing high-quality and relightable textures.

The diffusion model has been proven a powerful generative model in recent years, yet it remains a challenge in generating visual text. Although existing work has endeavored to enhance the accuracy of text rendering, these methods still suffer from several drawbacks, such as (1) limited flexibility and automation, (2) constrained capability of layout prediction, and (3) restricted diversity. In this paper, we present TextDiffuser-2, aiming to unleash the power of language models for text rendering while taking these three aspects into account. Firstly, we fine-tune a large language model for layout planning. The large language model is capable of automatically generating keywords and placing the text in optimal positions for text rendering. Secondly, we utilize the language model within the diffusion model to encode the position and content of keywords at the line level. Unlike previous methods that employed tight character-level guidance, our approach generates more diverse text images. We conduct extensive experiments and incorporate user studies involving human participants and GPT-4V, validating TextDiffuser-2's capacity to achieve a more rational text layout and generation with enhanced diversity. Furthermore, the proposed methods are compatible with existing text rendering techniques, such as TextDiffuser and GlyphControl, serving to enhance automation and diversity, as …

In this paper, we introduce a Multimodal Large Language Model-based Generation Assistant (LLMGA), leveraging the vast reservoir of knowledge and proficiency in reasoning, comprehension, and response inherent in Large Language Models (LLMs) to assist users in image generation and editing. Diverging from existing approaches where Multimodal Large Language Models (MLLMs) generate fixed-size embeddings to control Stable Diffusion (SD), our LLMGA provides a detailed language generation prompt for precise control over SD. This not only augments LLM context understanding but also reduces noise in generation prompts, yields images with more intricate and precise content, and elevates the interpretability of the network. To this end, we curate a comprehensive dataset comprising prompt refinement, similar image generation, inpainting \& outpainting, and instruction-based editing. Moreover, we propose a two-stage training scheme. In the first stage, we train the MLLM to grasp the properties of image generation and editing, enabling it to generate detailed prompts. In the second stage, we optimize SD to align with the MLLM's generation prompts. Additionally, we propose a reference-based restoration network to alleviate texture, brightness, and contrast disparities between generated and preserved regions during inpainting and outpainting. Extensive results show that LLMGA has promising generation and editing capabilities and can …

Diffusion models have significantly advanced the state of the art in image, audio, and video generation tasks. However, their applications in practical scenarios are hindered by slow inference speed. Drawing inspiration from the approximation strategies utilized in consistency models, we propose the Sub-path Linear Approximation Model (SLAM), which accelerates diffusion models while maintaining high-quality image generation. SLAM treats the PF-ODE trajectory as a series of PF-ODE sub-paths divided by sampled points, and harnesses sub-path linear (SL) ODEs to form a progressive and continuous error estimation along %the endpoints of each individual PF-ODE sub-path. The optimization on such SL-ODEs allows SLAM to construct denoising mappings with smaller cumulative approximated errors. An efficient distillation method is also developed to facilitate the incorporation of more advanced diffusion models, such as latent diffusion models. Our extensive experimental results demonstrate that SLAM achieves an efficient training regimen, requiring only 6 A100 GPU days to produce a high-quality generative model capable of 2 to 4-step generation with high performance. Comprehensive evaluations on LAION, MS COCO 2014, and MS COCO 2017 datasets also illustrate that SLAM surpasses existing acceleration methods in few-step generation tasks, achieving state-of-the-art performance both on FID and the quality of the generated images.

While recent advances in 3D-aware Generative Adversarial Networks (GANs) have aided the development of near-frontal view human face synthesis, the challenge of comprehensively synthesizing a full 3D head viewable from all angles still persists. Although PanoHead proves the possibilities of using a large-scale dataset with images of both frontal and back views for full-head synthesis, it often causes artifacts for back views. Based on our in-depth analysis, we found the reasons are mainly twofold. First, from network architecture perspective, we found each plane in the utilized tri-plane/tri-grid representation space tends to confuse the features from both sides, causing mirroring'' artifacts (e.g., the glasses appear in the back). Second, from data supervision aspect, we found that existing discriminator training in 3D GANs only focuses on the quality of the rendered image itself, and does not care about its plausibility with the perspective from which it was rendered. This makes it possible to generateface'' in the non-frontal view, due to its easiness to fool the discriminator. In response, we propose SphereHead, a novel tri-plane representation in the spherical coordinate system that fits the human head's geometric characteristics and efficiently mitigates many of the generated artifacts. We further introduce a view-image consistency …

Creating photorealistic avatars for individuals traditionally involves extensive capture sessions with complex and expensive studio devices like the LightStage system. While recent strides in neural representations have enabled the generation of photorealistic and animatable 3D avatars from quick phone scans, they have baked-in capture-time lighting, lack facial details and have missing regions in areas such as the back of the ears. Thus, they lag in quality compared to studio-captured avatars. In this paper, we propose a method to bridge this gap by generating studio-like illuminated texture maps from short, monocular phone captures. We do this by parameterizing the phone texture maps using the W+ space of a StyleGAN2, enabling near-perfect reconstruction. Then, we finetune a StyleGAN2 by sampling in the W+ parameterized space using a very small set of studio-captured textures as an adversarial training signal. To further enhance the realism and accuracy of facial details, we super-resolve the output of the StyleGAN2 using a diffusion model conditioned on image gradients of the phone-captured texture map. Once trained, our method excels at producing studio-like facial texture maps from casual monocular smartphone videos. Demonstrating its capabilities, we showcase the generation of photorealistic, uniformly lit, complete avatars from monocular phone captures.

Detecting AI-generated images has become an extraordinarily difficult challenge as new generative architectures emerge on a daily basis with more and more capabilities and unprecedented realism. New versions of many commercial tools, such as DALL·E, Midjourney, and Stable Diffusion, have been released recently, and it is impractical to continually update and retrain supervised forensic detectors to handle such a large variety of models. To address this challenge, we propose a zero-shot entropy-based detection method (ZSdet) that neither needs AI-generated training data nor relies on knowledge of generative architectures to artificially synthesize their artifacts. Inspired by recent works on machine-generated text detection, our idea is to measure how surprising the image under analysis is compared to a model of real images. To this end, we rely on a lossless image encoder that is able to estimate the probability distribution of each pixel given its context. To ensure computational efficiency, the encoder has a multi-resolution architecture and contexts comprise mostly pixels of the lower-resolution version of the image. Since only real images are needed to learn the model, the detector is independent of generator architectures and synthetic training data. Using a single discriminative feature, the proposed detector achieves state-of-the-art performance. On a …

Generating realistic audio for human interactions is important for many applications, such as creating sound effects for films or virtual reality games. Existing approaches implicitly assume total correspondence between the video and audio during training, yet many sounds happen off-screen and have weak to no correspondence with the visuals---resulting in uncontrolled ambient sounds or hallucinations at test time. We propose a novel ambient-aware audio generation model, AV-LDM. We devise a novel audio-conditioning mechanism to learn to disentangle foreground action sounds from the ambient background sounds in in-the-wild training videos. Given a novel silent video, our model uses retrieval-augmented generation to create audio that matches the visual content both semantically and temporally. We train and evaluate our model on two in-the-wild egocentric video datasets Ego4D and EPIC-KITCHENS. Our model outperforms an array of existing methods, allows controllable generation of the ambient sound, and even shows promise for generalizing to computer graphics game clips. Overall, our work is the first to focus video-to-audio generation faithfully on the observed visual content despite training from uncurated clips with natural background sounds.

Deep neural networks trained via empirical risk minimisation often exhibit significant performance disparities across groups, particularly when group and task labels are spuriously correlated (e.g., grassy background'' andcows''). Existing bias mitigation methods that aim to address this issue often either rely on group labels for training or validation, or require an extensive hyperparameter search. Such data and computational requirements hinder the practical deployment of these methods, especially when datasets are too large to be group-annotated, computational resources are limited, and models are trained through already complex pipelines. In this paper, we propose Targeted Augmentations for Bias Mitigation (TAB), a simple hyperparameter-free framework that leverages the entire training history of a helper model to identify spurious samples, and generate a group-balanced training set from which a robust model can be trained. We show that TAB improves worst-group performance without any group information or model selection, outperforming existing methods while maintaining overall accuracy.

Shortcut learning is when a model -- e.g. a cardiac disease classifier -- exploits correlations between the target label and a spurious shortcut feature, e.g. a pacemaker, to predict the target label based on the shortcut rather than real discriminative features. This is common in medical imaging, where treatment and clinical annotations correlate with disease labels, making them easy shortcuts to predict disease. We propose a novel detection and quantification of the impact of potential shortcut features via diffusion-based counterfactual image generation that can synthetically remove or add shortcuts. Via a novel self-optimized masking scheme we spatially limit the changes made with no extra inference step, encouraging the removal of spatially constrained shortcut features while ensuring that the shortcut-free counterfactuals preserve their remaining image features to a high degree. Using these, we assess how shortcut features influence model predictions. This is enabled by our second contribution: An efficient diffusion-based counterfactual generation with significant inference speed-up at comparable counterfactual quality as state-of-the-art. We confirm our performance on two large chest X-ray datasets, a skin lesion dataset, and CelebA.

We present the latest generation of MobileNets, known as MobileNetV4 (MNv4), featuring universally efficient architecture designs for mobile devices. At its core, we introduce the Universal Inverted Bottleneck (UIB) search block, a unified and flexible structure that merges Inverted Bottleneck (IB), ConvNext, Feed Forward Network (FFN), and a novel Extra Depthwise (ExtraDW) variant. Alongside UIB, we present Mobile MQA, an attention block tailored for mobile accelerators, delivering a significant 39% speedup. An optimized neural architecture search (NAS) recipe was also crafted to improve MNv4 search effectiveness. The integration of UIB, Mobile MQA and the refined NAS recipe results in a new suite of MNv4 models that are mostly Pareto optimal across mobile CPUs, DSPs, GPUs, as well as specialized accelerators like Apple Neural Engine and Google Pixel EdgeTPU - a characteristic not found in any other models tested. Our approach emphasizes simplicity, utilizing standard components and a straightforward attention mechanism to ensure broad hardware compatibility. To further boost efficiency, we finally introduce a novel distillation technique. Enhanced by this technique, our MNv4-Hybrid-Large model delivers impressive 87% ImageNet-1K accuracy, with a Pixel 8 EdgeTPU runtime of just 3.8ms.

Deep neural networks conventionally employ end-to-end backpropagation for their training process, which lacks biological credibility and triggers a locking dilemma during network parameter updates, leading to significant GPU memory use. Supervised local learning, which segments the network into multiple local blocks updated by independent auxiliary networks. However, these methods cannot replace end-to-end training due to lower accuracy, as gradients only propagate within their local block, creating a lack of information exchange between blocks. To address this issue and establish information transfer across blocks, we propose a Momentum Auxiliary Network (MAN) that establishes a dynamic interaction mechanism. The MAN leverages an exponential moving average (EMA) of the parameters from adjacent local blocks to enhance information flow. This auxiliary network, updated through EMA, helps bridge the informational gap between blocks. Nevertheless, we observe that directly applying EMA parameters has certain limitations due to feature discrepancies among local blocks. To overcome this, we introduce learnable biases, further boosting performance. We have validated our method on four image classification datasets (CIFAR-10, STL-10, SVHN, ImageNet), attaining superior performance and substantial memory savings. Notably, our method can reduce GPU memory usage by more than 45% on the ImageNet dataset compared to end-to-end training, while achieving higher …

We present a method for expanding a dataset by incorporating knowledge from the wide distribution of pre-trained latent diffusion models. Data augmentations typically incorporate inductive biases about the image formation process into the training (e.g. translation, scaling, colour changes, etc.). Here, we go beyond simple pixel transformations and introduce the concept of instance-level data augmentation by repainting parts of the image at the level of object instances. The method combines a conditional diffusion model with depth and edge maps control conditioning to seamlessly repaint individual objects inside the scene, being applicable to any segmentation or detection dataset. Used as a data augmentation method, it improves the performance and generalization of the state-of-the-art salient object detection, semantic segmentation and object detection models. By redrawing all privacy-sensitive instances (people, license plates, etc.), the method is also applicable for data anonymization. We also release fully synthetic and anonymized expansions for popular datasets: COCO, Pascal VOC and DUTS. All datasets and the code will be released.

The activation function plays a crucial role in model optimisation, yet the optimal choice remains unclear. For example, the Sigmoid activation is the de-facto activation in balanced classification tasks, however, in imbalanced classification, it proves inappropriate due to bias towards frequent classes. In this work, we delve deeper in this phenomenon by performing a comprehensive statistical analysis in the classification and intermediate layers of both balanced and imbalanced networks and we empirically show that aligning the activation function with the data distribution, enhances the performance in both balanced and imbalanced tasks. To this end, we propose the Adaptive Parametric Activation (APA) function, a novel and versatile activation function that unifies most common activation functions under a single formula. APA can be applied in both intermediate layers and attention layers, significantly outperforming the state-of-the-art on several imbalanced benchmarks such as ImageNet-LT, iNaturalist2018, Places-LT, CIFAR100-LT and LVIS and balanced benchmarks such as ImageNet1K, COCO and V3DET. Code is provided in the supplementary material.

This paper presents a general scheme for enhancing the convergence and performance of DETR (DEtection TRansformer). We investigate the slow convergence problem in Transformers from a new perspective, suggesting that it arises from the self-attention that introduces no structural bias over inputs. To address this issue, we explore incorporating position relation prior as attention bias to augment object detection, following the verification of its statistical significance using a proposed quantitative macrosopic correlation (MC) metric. Our approach, termed Relation-DETR, introduces an encoder to construct position relation embeddings for progressive attention refinment, which further extends the traditional streaming pipeline of DETR into a contrastive relation pipeline to address the conflicts between non-duplicate predictions and positive supervision. Extensive experiments on both generic and task-specific datasets demonstrate the effectiveness of our approach. Under the same configurations, Relation-DETR achieves a significant improvement (+2.0% AP) on COCO 2017 compared to DINO, the previous highly optimized DETR detector, and achieves state-of-the-art performance (reaching 51.7% at 12 epochs and 52.1% AP at 24 epochs with a ResNet-50 backbone, respectively). Moreover, Relation-DETR exhibits a remarkably fast convergence speed, achieving over 40% AP with only 2 training epochs on COCO 2017 using the basic ResNet50 backbone, suppressing existing DETR detectors …

This paper introduces the Point-Axis representation for oriented objects in aerial images, as depicted in Figure 1, emphasizing its flexibility and geometrically intuitive nature with two key components: points and axes. 1) Points delineate the spatial extent and contours of objects, providing detailed shape descriptions. 2) Axes define the primary directionalities of objects, providing essential orientation cues crucial for precise detection. The point-axis representation decouples location and rotation, addressing the loss discontinuity issues commonly encountered in traditional bounding box based approaches. For effective optimization without introducing additional annotations, we propose the max-projection loss to supervise point set learning and the cross-axis loss for robust axis representation learning. Further, leveraging this representation, we present the Oriented DETR model, seamlessly integrating the DETR framework for precise point-axis prediction and end-to-end detection. Experimental results demonstrate effectiveness in main datasets, showing significant performance improvements in aerial-oriented object detection tasks. The code will be released to the community.

Open-vocabulary object detection (OVD) utilizes image-level cues to expand the linguistic space of region proposals, thereby facilitating the detection of diverse novel classes. Recent works adapt CLIP embedding by minimizing the object-image and object-text discrepancy combinatorially in a discriminative paradigm. However, they ignore the underlying distribution and the disagreement between the image and text objective, leading to the misaligned distribution between the vision and language sub-space. To address the deficiency, we explore the advanced generative paradigm with distribution perception and propose a novel framework based on the diffusion model, coined Continual Latent Diffusion (CLIFF), which formulates a continual distribution transfer among the object, image, and text latent space probabilistically. CLIFF consists of a Variational Latent Sampler (VLS) enabling the probabilistic modeling and a Continual Diffusion Module (CDM) for the distribution transfer. Specifically, in VLS, we first establish a probabilistic object space with region proposals by estimating distribution parameters. Then, the object-centric noise is sampled from the estimated distribution to generate text embedding for OVD. To achieve this generation process, CDM conducts a short-distance object-to-image diffusion from the sampled noise to generate image embedding as the medium, which guides the long-distance diffusion to generate text embedding. Extensive experiments verify that CLIFF …

Building calibrated object detectors is a crucial challenge to address for their reliable usage in safety-critical applications. Recent approaches towards this involve (1) designing new loss functions to obtain calibrated detectors by training them from scratch, and (2) post-hoc Temperature Scaling (TS) that learns to scale the likelihood of a trained detector to output calibrated predictions. These approaches are then evaluated based on a combination of Detection Expected Calibration Error (D-ECE) and Average Precision. In this work, via extensive analysis and insights, we highlight that these recent evaluation frameworks, evaluation metrics, and the use of TS have significant drawbacks leading to incorrect conclusions. As a remedy, we propose a principled evaluation framework to jointly measure calibration and accuracy of object detectors. We also tailor efficient and easy-to-use post-hoc calibration approaches, Platt Scaling and Isotonic Regression, specifically to object detection. As opposed to the common notion, our experiments show that, once designed and evaluated properly, post-hoc calibrators, which are extremely cheap to build and use, are much more powerful and effective than the recent train time calibration methods. To illustrate, D-DETR with our post-hoc Isotonic Regression calibrator outperforms the state-of-the-art Cal-DETR by more than 7 D-ECE on the COCO dataset. We …

In this paper, we present a groundbreaking spectrally multiplexed photometric stereo approach for recovering surface normals of dynamic surfaces without the need for calibrated lighting or sensors, a notable advancement in the field traditionally hindered by stringent prerequisites and spectral ambiguity. By embracing spectral ambiguity as an advantage, our technique enables the generation of training data without specialized multispectral rendering frameworks. We introduce a unique, physics-free network architecture, SpectraM-PS, that effectively processes multiplexed images to determine surface normals across a wide range of conditions and material types, without relying on specific physically-based knowledge. Additionally, we establish the first benchmark dataset, SpectraM14, for spectrally multiplexed photometric stereo, facilitating comprehensive evaluations against existing calibrated methods. Our contributions significantly enhance the capabilities for dynamic surface recovery, particularly in uncalibrated setups, marking a pivotal step forward in the application of photometric stereo across various domains.

We present COMO, a real-time monocular mapping and odometry system that encodes dense geometry via a compact set of 3D anchor points. Decoding anchor point projections into dense geometry via per-keyframe depth covariance functions guarantees that depth maps are joined together at visible anchor points. The representation enables joint optimization of camera poses and dense geometry, intrinsic 3D consistency, and efficient second-order inference. To maintain a compact yet expressive map, we introduce a frontend that leverages the covariance function for tracking and initializing potentially visually indistinct 3D points across frames. Altogether, we introduce a real-time system capable of estimating accurate poses and consistent geometry.

Despite significant progress in unsupervised multi-view stereo (MVS), the core loss formulation has remained largely unchanged since its introduction. However, we identify fundamental limitations to this core loss and propose three major changes to improve the modeling of depth priors, occlusion, and view-dependent effects. First, we eliminate prominent stair-stepping and edge artifacts in predicted depth maps using a clamped depth-smoothness constraint. Second, we propose a learned view-synthesis approach to generate an image for photometric loss, avoiding the use of hand-coded heuristics for handling view-dependent effects. Finally, we sample additional views for supervision beyond those used as MVS input, challenging the network to predict depth that matches unseen views. Together, these contributions form an improved supervision strategy we call DIV loss. The key advantage of our DIV loss is that it can be easily dropped into existing unsupervised MVS training pipelines, resulting in significant improvements on competitive reconstruction benchmarks and drastically better qualitative performance around object boundaries for minimal training cost.

Recovering camera poses from a set of images is a foundational task in 3D computer vision, which powers key applications such as 3D scene/object reconstructions. Classic methods often depend on feature correspondence, such as keypoints, which require the input images to have large overlap and small viewpoint changes. Such requirements present considerable challenges in scenarios with sparse views. Recent data-driven approaches aim to directly output camera poses, either through regressing the 6DoF camera poses or formulating rotation as a probability distribution. However, each approach has its limitations. On one hand, directly regressing the camera poses can be ill-posed, since it assumes a single mode, which is not true under symmetry and leads to sub-optimal solutions. On the other hand, probabilistic approaches are capable of modeling the symmetry ambiguity, yet they sample the entire space of rotation uniformly by brute-force. This leads to an inevitable trade-off between high sample density, which improves model precision, and sample efficiency that determines the runtime. In this paper, we propose ADen to unify the two frameworks by employing a generator and a discriminator: the generator is trained to output multiple hypotheses of 6DoF camera pose to represent a distribution and handle multi-mode ambiguity, and the …

In this paper, we present SPVLoc, a global indoor localization method that accurately determines the six-dimensional (6D) camera pose of a query image and requires minimal scene-specific prior knowledge and no scene-specific training. Our approach employs a novel matching procedure to localize the perspective camera's viewport, given as an RGB image, within a set of panoramic semantic layout representations of the indoor environment. The panoramas are rendered from an untextured 3D reference model, which only compromises approximate structural information about room shapes, along with door and window annotations. We demonstrate that a straightforward convolutional network structure can successfully achieve image-to-panorama and ultimately image-to-model matching. Through a viewport classification score, we rank reference panoramas and select the best match for the query image. Then, a 6D relative pose is estimated between the chosen panorama and query image. Our experiments demonstrate that this approach not only efficiently bridges the domain gap but also generalizes well to previously unseen scenes that are not part of the training data. Moreover, it achieves superior localization accuracy compared to the state of the art methods and also estimates more degrees of freedom of the camera pose.

Relative pose estimation using point correspondences (PC) is a widely used technique. A minimal configuration of six PCs is required for generalized cameras. In this paper, we present several minimal solvers that use six PCs to compute the 6DOF relative pose of a multi-camera system, including a minimal solver for the generalized camera and two minimal solvers for the practical configuration of two-camera rigs. The equation construction is based on the decoupling of rotation and translation. Rotation is represented by Cayley or quaternion parametrization, and translation can be eliminated by using the hidden variable technique. Ray bundle constraints are found and proven when a subset of PCs relate the same cameras across two views. This is the key to reducing the number of solutions and generating numerically stable solvers. Moreover, all configurations of six-point problems for multi-camera systems are enumerated. Extensive experiments demonstrate that our solvers are more accurate than the state-of-the-art six-point methods, while achieving better performance in efficiency.

We address the task of estimating camera parameters from a set of images depicting a scene. Popular feature-based structure-from-motion (SfM) tools solve this task by incremental reconstruction: they repeat triangulation of sparse 3D points and registration of more camera views to the sparse point cloud. We re-interpret incremental structure-from-motion as an iterated application and refinement of a visual relocalizer, that is, of a method that registers new views to the current state of the reconstruction. This perspective allows us to investigate alternative visual relocalizers that are not rooted in local feature matching. We show that scene coordinate regression, a learning-based relocalization approach, allows us to build implicit, neural scene representations from unposed images. Different from other learning-based reconstruction methods, we do not require pose priors nor sequential inputs, and we optimize efficiently over thousands of images. Our method, ACE0, estimates camera poses to an accuracy comparable to feature-based SfM, as demonstrated by novel view synthesis.

Image Matching is a core component of all best-performing algorithms and pipelines in 3D vision. Yet despite matching being fundamentally a 3D problem, intrinsically linked to camera pose and scene geometry, it is typically treated as a 2D problem. This makes sense as the goal of matching is to establish correspondences between 2D pixel fields, but also seems like a potentially hazardous choice. In this work, we take a different stance and propose to cast matching as a 3D task with DUSt3R, a recent and powerful 3D reconstruction framework based on Transformers. Based on pointmaps regression, this method displayed impressive robustness in matching views with extreme viewpoint changes, yet with limited accuracy. We aim here to improve the matching capabilities of such an approach while preserving its robustness. We thus propose to augment the DUSt3R network with a new head that outputs dense local features, trained with an additional matching loss. We further address the issue of quadratic complexity of dense matching, which becomes prohibitively slow for downstream applications if not treated carefully. We introduce a fast reciprocal matching scheme that not only accelerates matching by orders of magnitude, but also comes with theoretical guarantees and, lastly, yields improved results. …

To advance the state of the art in the creation of 3D foundation models, this paper introduces the ConDense framework for 3D pre-training utilizing existing pre-trained 2D networks and large-scale multi-view datasets. We propose a novel 2D-3D joint training scheme to extract co-embedded 2D and 3D features in an end-to-end pipeline, where 2D-3D feature consistency is enforced through a volume rendering NeRF-like ray marching process. Using dense per pixel features we are able to 1) directly distill the learned priors from 2D models to 3D models and create useful 3D backbones, 2) extract more consistent and less noisy 2D features, 3) formulate a consistent embedding space where 2D, 3D, and other modalities of data (e.g., natural language prompts) can be jointly queried. Furthermore, besides dense features, ConDense can be trained to extract sparse features (e.g., key points), also with 2D-3D consistency -- condensing 3D NeRF representations into compact sets of decorated key points. We demonstrate that our pre-trained model provides good initialization for various 3D tasks including 3D classification and segmentation, outperforming other 3D pre-training methods by a significant margin. It also enables, by exploiting our sparse features, additional useful downstream tasks, such as matching 2D images to 3D scenes, …

Computer vision has long relied on two kinds of correspondences: pixel correspondences in images and 3D correspondences on object surfaces. Is there another kind, and if there is, what can they do for us? In this paper, we introduce correspondences of the third kind we call reflection correspondences and show that they can help estimate camera pose by just looking at objects without relying on the background. Reflection correspondences are point correspondences in the reflected world, i.e., the scene reflected by the object surface. The object geometry and reflectance alters the scene geometrically and radiometrically, respectively, causing incorrect pixel correspondences. Geometry recovered from each image is also hampered by distortions, namely generalized bas-relief ambiguity, leading to erroneous 3D correspondences. We show that reflection correspondences can resolve the ambiguities arising from these distortions. We introduce a neural correspondence estimator and a RANSAC algorithm that fully leverages all three kinds of correspondences for robust and accurate joint camera pose and object shape estimation just from the object appearance. The method expands the horizon of numerous downstream tasks, including camera pose estimation for appearance modeling (e.g., NeRF) and motion estimation of reflective objects (e.g., cars on the road), to name a few, as …

Camera calibration is a crucial step in photogrammetry and 3D vision applications. In practical application scenarios where the camera has a long working distance to cover a wide area, target-based calibration methods become complicated and inflexible due to site limitations. This paper introduces a novel camera calibration method using a designed collimator system. This allows the camera to clearly observe the calibration pattern at a close range. Based on the optical geometry of collimator system, we prove that the motion of the calibration target conforms to the spherical motion model with respect to the camera. The spherical motion constraint reduces the original 6DOF motion to 3DOF pure rotation. Moreover, a closed-form solver for multiple views and a minimal solver for two views are proposed for camera calibration. The performance of the proposed method is tested in both synthetic and real-world experiments, which demonstrates that the calibration accuracy is superior to the state-of-the-art methods.

Detecting AI-generated images has become an extraordinarily difficult challenge as new generative architectures emerge on a daily basis with more and more capabilities and unprecedented realism. New versions of many commercial tools, such as DALL·E, Midjourney, and Stable Diffusion, have been released recently, and it is impractical to continually update and retrain supervised forensic detectors to handle such a large variety of models. To address this challenge, we propose a zero-shot entropy-based detection method (ZSdet) that neither needs AI-generated training data nor relies on knowledge of generative architectures to artificially synthesize their artifacts. Inspired by recent works on machine-generated text detection, our idea is to measure how surprising the image under analysis is compared to a model of real images. To this end, we rely on a lossless image encoder that is able to estimate the probability distribution of each pixel given its context. To ensure computational efficiency, the encoder has a multi-resolution architecture and contexts comprise mostly pixels of the lower-resolution version of the image. Since only real images are needed to learn the model, the detector is independent of generator architectures and synthetic training data. Using a single discriminative feature, the proposed detector achieves state-of-the-art performance. On a …

We present the latest generation of MobileNets, known as MobileNetV4 (MNv4), featuring universally efficient architecture designs for mobile devices. At its core, we introduce the Universal Inverted Bottleneck (UIB) search block, a unified and flexible structure that merges Inverted Bottleneck (IB), ConvNext, Feed Forward Network (FFN), and a novel Extra Depthwise (ExtraDW) variant. Alongside UIB, we present Mobile MQA, an attention block tailored for mobile accelerators, delivering a significant 39% speedup. An optimized neural architecture search (NAS) recipe was also crafted to improve MNv4 search effectiveness. The integration of UIB, Mobile MQA and the refined NAS recipe results in a new suite of MNv4 models that are mostly Pareto optimal across mobile CPUs, DSPs, GPUs, as well as specialized accelerators like Apple Neural Engine and Google Pixel EdgeTPU - a characteristic not found in any other models tested. Our approach emphasizes simplicity, utilizing standard components and a straightforward attention mechanism to ensure broad hardware compatibility. To further boost efficiency, we finally introduce a novel distillation technique. Enhanced by this technique, our MNv4-Hybrid-Large model delivers impressive 87% ImageNet-1K accuracy, with a Pixel 8 EdgeTPU runtime of just 3.8ms.

Shortcut learning is when a model -- e.g. a cardiac disease classifier -- exploits correlations between the target label and a spurious shortcut feature, e.g. a pacemaker, to predict the target label based on the shortcut rather than real discriminative features. This is common in medical imaging, where treatment and clinical annotations correlate with disease labels, making them easy shortcuts to predict disease. We propose a novel detection and quantification of the impact of potential shortcut features via diffusion-based counterfactual image generation that can synthetically remove or add shortcuts. Via a novel self-optimized masking scheme we spatially limit the changes made with no extra inference step, encouraging the removal of spatially constrained shortcut features while ensuring that the shortcut-free counterfactuals preserve their remaining image features to a high degree. Using these, we assess how shortcut features influence model predictions. This is enabled by our second contribution: An efficient diffusion-based counterfactual generation with significant inference speed-up at comparable counterfactual quality as state-of-the-art. We confirm our performance on two large chest X-ray datasets, a skin lesion dataset, and CelebA.

The activation function plays a crucial role in model optimisation, yet the optimal choice remains unclear. For example, the Sigmoid activation is the de-facto activation in balanced classification tasks, however, in imbalanced classification, it proves inappropriate due to bias towards frequent classes. In this work, we delve deeper in this phenomenon by performing a comprehensive statistical analysis in the classification and intermediate layers of both balanced and imbalanced networks and we empirically show that aligning the activation function with the data distribution, enhances the performance in both balanced and imbalanced tasks. To this end, we propose the Adaptive Parametric Activation (APA) function, a novel and versatile activation function that unifies most common activation functions under a single formula. APA can be applied in both intermediate layers and attention layers, significantly outperforming the state-of-the-art on several imbalanced benchmarks such as ImageNet-LT, iNaturalist2018, Places-LT, CIFAR100-LT and LVIS and balanced benchmarks such as ImageNet1K, COCO and V3DET. Code is provided in the supplementary material.

Open-vocabulary object detection (OVD) utilizes image-level cues to expand the linguistic space of region proposals, thereby facilitating the detection of diverse novel classes. Recent works adapt CLIP embedding by minimizing the object-image and object-text discrepancy combinatorially in a discriminative paradigm. However, they ignore the underlying distribution and the disagreement between the image and text objective, leading to the misaligned distribution between the vision and language sub-space. To address the deficiency, we explore the advanced generative paradigm with distribution perception and propose a novel framework based on the diffusion model, coined Continual Latent Diffusion (CLIFF), which formulates a continual distribution transfer among the object, image, and text latent space probabilistically. CLIFF consists of a Variational Latent Sampler (VLS) enabling the probabilistic modeling and a Continual Diffusion Module (CDM) for the distribution transfer. Specifically, in VLS, we first establish a probabilistic object space with region proposals by estimating distribution parameters. Then, the object-centric noise is sampled from the estimated distribution to generate text embedding for OVD. To achieve this generation process, CDM conducts a short-distance object-to-image diffusion from the sampled noise to generate image embedding as the medium, which guides the long-distance diffusion to generate text embedding. Extensive experiments verify that CLIFF …

We present a method for expanding a dataset by incorporating knowledge from the wide distribution of pre-trained latent diffusion models. Data augmentations typically incorporate inductive biases about the image formation process into the training (e.g. translation, scaling, colour changes, etc.). Here, we go beyond simple pixel transformations and introduce the concept of instance-level data augmentation by repainting parts of the image at the level of object instances. The method combines a conditional diffusion model with depth and edge maps control conditioning to seamlessly repaint individual objects inside the scene, being applicable to any segmentation or detection dataset. Used as a data augmentation method, it improves the performance and generalization of the state-of-the-art salient object detection, semantic segmentation and object detection models. By redrawing all privacy-sensitive instances (people, license plates, etc.), the method is also applicable for data anonymization. We also release fully synthetic and anonymized expansions for popular datasets: COCO, Pascal VOC and DUTS. All datasets and the code will be released.

Deep neural networks trained via empirical risk minimisation often exhibit significant performance disparities across groups, particularly when group and task labels are spuriously correlated (e.g., grassy background'' andcows''). Existing bias mitigation methods that aim to address this issue often either rely on group labels for training or validation, or require an extensive hyperparameter search. Such data and computational requirements hinder the practical deployment of these methods, especially when datasets are too large to be group-annotated, computational resources are limited, and models are trained through already complex pipelines. In this paper, we propose Targeted Augmentations for Bias Mitigation (TAB), a simple hyperparameter-free framework that leverages the entire training history of a helper model to identify spurious samples, and generate a group-balanced training set from which a robust model can be trained. We show that TAB improves worst-group performance without any group information or model selection, outperforming existing methods while maintaining overall accuracy.

Building calibrated object detectors is a crucial challenge to address for their reliable usage in safety-critical applications. Recent approaches towards this involve (1) designing new loss functions to obtain calibrated detectors by training them from scratch, and (2) post-hoc Temperature Scaling (TS) that learns to scale the likelihood of a trained detector to output calibrated predictions. These approaches are then evaluated based on a combination of Detection Expected Calibration Error (D-ECE) and Average Precision. In this work, via extensive analysis and insights, we highlight that these recent evaluation frameworks, evaluation metrics, and the use of TS have significant drawbacks leading to incorrect conclusions. As a remedy, we propose a principled evaluation framework to jointly measure calibration and accuracy of object detectors. We also tailor efficient and easy-to-use post-hoc calibration approaches, Platt Scaling and Isotonic Regression, specifically to object detection. As opposed to the common notion, our experiments show that, once designed and evaluated properly, post-hoc calibrators, which are extremely cheap to build and use, are much more powerful and effective than the recent train time calibration methods. To illustrate, D-DETR with our post-hoc Isotonic Regression calibrator outperforms the state-of-the-art Cal-DETR by more than 7 D-ECE on the COCO dataset. We …

Deep neural networks conventionally employ end-to-end backpropagation for their training process, which lacks biological credibility and triggers a locking dilemma during network parameter updates, leading to significant GPU memory use. Supervised local learning, which segments the network into multiple local blocks updated by independent auxiliary networks. However, these methods cannot replace end-to-end training due to lower accuracy, as gradients only propagate within their local block, creating a lack of information exchange between blocks. To address this issue and establish information transfer across blocks, we propose a Momentum Auxiliary Network (MAN) that establishes a dynamic interaction mechanism. The MAN leverages an exponential moving average (EMA) of the parameters from adjacent local blocks to enhance information flow. This auxiliary network, updated through EMA, helps bridge the informational gap between blocks. Nevertheless, we observe that directly applying EMA parameters has certain limitations due to feature discrepancies among local blocks. To overcome this, we introduce learnable biases, further boosting performance. We have validated our method on four image classification datasets (CIFAR-10, STL-10, SVHN, ImageNet), attaining superior performance and substantial memory savings. Notably, our method can reduce GPU memory usage by more than 45% on the ImageNet dataset compared to end-to-end training, while achieving higher …

This paper introduces the Point-Axis representation for oriented objects in aerial images, as depicted in Figure 1, emphasizing its flexibility and geometrically intuitive nature with two key components: points and axes. 1) Points delineate the spatial extent and contours of objects, providing detailed shape descriptions. 2) Axes define the primary directionalities of objects, providing essential orientation cues crucial for precise detection. The point-axis representation decouples location and rotation, addressing the loss discontinuity issues commonly encountered in traditional bounding box based approaches. For effective optimization without introducing additional annotations, we propose the max-projection loss to supervise point set learning and the cross-axis loss for robust axis representation learning. Further, leveraging this representation, we present the Oriented DETR model, seamlessly integrating the DETR framework for precise point-axis prediction and end-to-end detection. Experimental results demonstrate effectiveness in main datasets, showing significant performance improvements in aerial-oriented object detection tasks. The code will be released to the community.

This paper presents a general scheme for enhancing the convergence and performance of DETR (DEtection TRansformer). We investigate the slow convergence problem in Transformers from a new perspective, suggesting that it arises from the self-attention that introduces no structural bias over inputs. To address this issue, we explore incorporating position relation prior as attention bias to augment object detection, following the verification of its statistical significance using a proposed quantitative macrosopic correlation (MC) metric. Our approach, termed Relation-DETR, introduces an encoder to construct position relation embeddings for progressive attention refinment, which further extends the traditional streaming pipeline of DETR into a contrastive relation pipeline to address the conflicts between non-duplicate predictions and positive supervision. Extensive experiments on both generic and task-specific datasets demonstrate the effectiveness of our approach. Under the same configurations, Relation-DETR achieves a significant improvement (+2.0% AP) on COCO 2017 compared to DINO, the previous highly optimized DETR detector, and achieves state-of-the-art performance (reaching 51.7% at 12 epochs and 52.1% AP at 24 epochs with a ResNet-50 backbone, respectively). Moreover, Relation-DETR exhibits a remarkably fast convergence speed, achieving over 40% AP with only 2 training epochs on COCO 2017 using the basic ResNet50 backbone, suppressing existing DETR detectors …

To advance the state of the art in the creation of 3D foundation models, this paper introduces the ConDense framework for 3D pre-training utilizing existing pre-trained 2D networks and large-scale multi-view datasets. We propose a novel 2D-3D joint training scheme to extract co-embedded 2D and 3D features in an end-to-end pipeline, where 2D-3D feature consistency is enforced through a volume rendering NeRF-like ray marching process. Using dense per pixel features we are able to 1) directly distill the learned priors from 2D models to 3D models and create useful 3D backbones, 2) extract more consistent and less noisy 2D features, 3) formulate a consistent embedding space where 2D, 3D, and other modalities of data (e.g., natural language prompts) can be jointly queried. Furthermore, besides dense features, ConDense can be trained to extract sparse features (e.g., key points), also with 2D-3D consistency -- condensing 3D NeRF representations into compact sets of decorated key points. We demonstrate that our pre-trained model provides good initialization for various 3D tasks including 3D classification and segmentation, outperforming other 3D pre-training methods by a significant margin. It also enables, by exploiting our sparse features, additional useful downstream tasks, such as matching 2D images to 3D scenes, …

Recovering camera poses from a set of images is a foundational task in 3D computer vision, which powers key applications such as 3D scene/object reconstructions. Classic methods often depend on feature correspondence, such as keypoints, which require the input images to have large overlap and small viewpoint changes. Such requirements present considerable challenges in scenarios with sparse views. Recent data-driven approaches aim to directly output camera poses, either through regressing the 6DoF camera poses or formulating rotation as a probability distribution. However, each approach has its limitations. On one hand, directly regressing the camera poses can be ill-posed, since it assumes a single mode, which is not true under symmetry and leads to sub-optimal solutions. On the other hand, probabilistic approaches are capable of modeling the symmetry ambiguity, yet they sample the entire space of rotation uniformly by brute-force. This leads to an inevitable trade-off between high sample density, which improves model precision, and sample efficiency that determines the runtime. In this paper, we propose ADen to unify the two frameworks by employing a generator and a discriminator: the generator is trained to output multiple hypotheses of 6DoF camera pose to represent a distribution and handle multi-mode ambiguity, and the …

We present COMO, a real-time monocular mapping and odometry system that encodes dense geometry via a compact set of 3D anchor points. Decoding anchor point projections into dense geometry via per-keyframe depth covariance functions guarantees that depth maps are joined together at visible anchor points. The representation enables joint optimization of camera poses and dense geometry, intrinsic 3D consistency, and efficient second-order inference. To maintain a compact yet expressive map, we introduce a frontend that leverages the covariance function for tracking and initializing potentially visually indistinct 3D points across frames. Altogether, we introduce a real-time system capable of estimating accurate poses and consistent geometry.

Camera calibration is a crucial step in photogrammetry and 3D vision applications. In practical application scenarios where the camera has a long working distance to cover a wide area, target-based calibration methods become complicated and inflexible due to site limitations. This paper introduces a novel camera calibration method using a designed collimator system. This allows the camera to clearly observe the calibration pattern at a close range. Based on the optical geometry of collimator system, we prove that the motion of the calibration target conforms to the spherical motion model with respect to the camera. The spherical motion constraint reduces the original 6DOF motion to 3DOF pure rotation. Moreover, a closed-form solver for multiple views and a minimal solver for two views are proposed for camera calibration. The performance of the proposed method is tested in both synthetic and real-world experiments, which demonstrates that the calibration accuracy is superior to the state-of-the-art methods.

Computer vision has long relied on two kinds of correspondences: pixel correspondences in images and 3D correspondences on object surfaces. Is there another kind, and if there is, what can they do for us? In this paper, we introduce correspondences of the third kind we call reflection correspondences and show that they can help estimate camera pose by just looking at objects without relying on the background. Reflection correspondences are point correspondences in the reflected world, i.e., the scene reflected by the object surface. The object geometry and reflectance alters the scene geometrically and radiometrically, respectively, causing incorrect pixel correspondences. Geometry recovered from each image is also hampered by distortions, namely generalized bas-relief ambiguity, leading to erroneous 3D correspondences. We show that reflection correspondences can resolve the ambiguities arising from these distortions. We introduce a neural correspondence estimator and a RANSAC algorithm that fully leverages all three kinds of correspondences for robust and accurate joint camera pose and object shape estimation just from the object appearance. The method expands the horizon of numerous downstream tasks, including camera pose estimation for appearance modeling (e.g., NeRF) and motion estimation of reflective objects (e.g., cars on the road), to name a few, as …

In this paper, we present a groundbreaking spectrally multiplexed photometric stereo approach for recovering surface normals of dynamic surfaces without the need for calibrated lighting or sensors, a notable advancement in the field traditionally hindered by stringent prerequisites and spectral ambiguity. By embracing spectral ambiguity as an advantage, our technique enables the generation of training data without specialized multispectral rendering frameworks. We introduce a unique, physics-free network architecture, SpectraM-PS, that effectively processes multiplexed images to determine surface normals across a wide range of conditions and material types, without relying on specific physically-based knowledge. Additionally, we establish the first benchmark dataset, SpectraM14, for spectrally multiplexed photometric stereo, facilitating comprehensive evaluations against existing calibrated methods. Our contributions significantly enhance the capabilities for dynamic surface recovery, particularly in uncalibrated setups, marking a pivotal step forward in the application of photometric stereo across various domains.

In this paper, we present SPVLoc, a global indoor localization method that accurately determines the six-dimensional (6D) camera pose of a query image and requires minimal scene-specific prior knowledge and no scene-specific training. Our approach employs a novel matching procedure to localize the perspective camera's viewport, given as an RGB image, within a set of panoramic semantic layout representations of the indoor environment. The panoramas are rendered from an untextured 3D reference model, which only compromises approximate structural information about room shapes, along with door and window annotations. We demonstrate that a straightforward convolutional network structure can successfully achieve image-to-panorama and ultimately image-to-model matching. Through a viewport classification score, we rank reference panoramas and select the best match for the query image. Then, a 6D relative pose is estimated between the chosen panorama and query image. Our experiments demonstrate that this approach not only efficiently bridges the domain gap but also generalizes well to previously unseen scenes that are not part of the training data. Moreover, it achieves superior localization accuracy compared to the state of the art methods and also estimates more degrees of freedom of the camera pose.

Despite significant progress in unsupervised multi-view stereo (MVS), the core loss formulation has remained largely unchanged since its introduction. However, we identify fundamental limitations to this core loss and propose three major changes to improve the modeling of depth priors, occlusion, and view-dependent effects. First, we eliminate prominent stair-stepping and edge artifacts in predicted depth maps using a clamped depth-smoothness constraint. Second, we propose a learned view-synthesis approach to generate an image for photometric loss, avoiding the use of hand-coded heuristics for handling view-dependent effects. Finally, we sample additional views for supervision beyond those used as MVS input, challenging the network to predict depth that matches unseen views. Together, these contributions form an improved supervision strategy we call DIV loss. The key advantage of our DIV loss is that it can be easily dropped into existing unsupervised MVS training pipelines, resulting in significant improvements on competitive reconstruction benchmarks and drastically better qualitative performance around object boundaries for minimal training cost.

Relative pose estimation using point correspondences (PC) is a widely used technique. A minimal configuration of six PCs is required for generalized cameras. In this paper, we present several minimal solvers that use six PCs to compute the 6DOF relative pose of a multi-camera system, including a minimal solver for the generalized camera and two minimal solvers for the practical configuration of two-camera rigs. The equation construction is based on the decoupling of rotation and translation. Rotation is represented by Cayley or quaternion parametrization, and translation can be eliminated by using the hidden variable technique. Ray bundle constraints are found and proven when a subset of PCs relate the same cameras across two views. This is the key to reducing the number of solutions and generating numerically stable solvers. Moreover, all configurations of six-point problems for multi-camera systems are enumerated. Extensive experiments demonstrate that our solvers are more accurate than the state-of-the-art six-point methods, while achieving better performance in efficiency.

We address the task of estimating camera parameters from a set of images depicting a scene. Popular feature-based structure-from-motion (SfM) tools solve this task by incremental reconstruction: they repeat triangulation of sparse 3D points and registration of more camera views to the sparse point cloud. We re-interpret incremental structure-from-motion as an iterated application and refinement of a visual relocalizer, that is, of a method that registers new views to the current state of the reconstruction. This perspective allows us to investigate alternative visual relocalizers that are not rooted in local feature matching. We show that scene coordinate regression, a learning-based relocalization approach, allows us to build implicit, neural scene representations from unposed images. Different from other learning-based reconstruction methods, we do not require pose priors nor sequential inputs, and we optimize efficiently over thousands of images. Our method, ACE0, estimates camera poses to an accuracy comparable to feature-based SfM, as demonstrated by novel view synthesis.

Image Matching is a core component of all best-performing algorithms and pipelines in 3D vision. Yet despite matching being fundamentally a 3D problem, intrinsically linked to camera pose and scene geometry, it is typically treated as a 2D problem. This makes sense as the goal of matching is to establish correspondences between 2D pixel fields, but also seems like a potentially hazardous choice. In this work, we take a different stance and propose to cast matching as a 3D task with DUSt3R, a recent and powerful 3D reconstruction framework based on Transformers. Based on pointmaps regression, this method displayed impressive robustness in matching views with extreme viewpoint changes, yet with limited accuracy. We aim here to improve the matching capabilities of such an approach while preserving its robustness. We thus propose to augment the DUSt3R network with a new head that outputs dense local features, trained with an additional matching loss. We further address the issue of quadratic complexity of dense matching, which becomes prohibitively slow for downstream applications if not treated carefully. We introduce a fast reciprocal matching scheme that not only accelerates matching by orders of magnitude, but also comes with theoretical guarantees and, lastly, yields improved results. …

Achieving disentangled control over multiple facial motions and accommodating diverse input modalities greatly enhances the application and entertainment of the talking head generation. This necessitates a deep exploration of the decoupling space for facial features, ensuring that they a) operate independently without mutual interference and b) can be preserved to share with different modal inputs—both aspects often neglected in existing methods. To address this gap, this paper proposes a novel Efficient Disentanglement framework for Talking head generation (EDTalk). Our framework enables individual manipulation of mouth shape, head pose, and emotional expression, conditioned on video or audio inputs. Specifically, we employ three lightweight modules to decompose the facial dynamics into three distinct latent spaces representing mouth, pose, and expression, respectively. Each space is characterized by a set of learnable bases whose linear combinations define specific motions. To ensure independence and accelerate training, we enforce orthogonality among bases and devise an efficient training strategy to allocate motion responsibilities to each space without relying on external knowledge. The learned bases are then stored in corresponding banks, enabling shared visual priors with audio input. Furthermore, considering the properties of each space, we propose an Audio-to-Motion module for audio-driven talking head synthesis. Experiments are conducted …

The diffusion model has been proven a powerful generative model in recent years, yet it remains a challenge in generating visual text. Although existing work has endeavored to enhance the accuracy of text rendering, these methods still suffer from several drawbacks, such as (1) limited flexibility and automation, (2) constrained capability of layout prediction, and (3) restricted diversity. In this paper, we present TextDiffuser-2, aiming to unleash the power of language models for text rendering while taking these three aspects into account. Firstly, we fine-tune a large language model for layout planning. The large language model is capable of automatically generating keywords and placing the text in optimal positions for text rendering. Secondly, we utilize the language model within the diffusion model to encode the position and content of keywords at the line level. Unlike previous methods that employed tight character-level guidance, our approach generates more diverse text images. We conduct extensive experiments and incorporate user studies involving human participants and GPT-4V, validating TextDiffuser-2's capacity to achieve a more rational text layout and generation with enhanced diversity. Furthermore, the proposed methods are compatible with existing text rendering techniques, such as TextDiffuser and GlyphControl, serving to enhance automation and diversity, as …

Diffusion models have significantly advanced the state of the art in image, audio, and video generation tasks. However, their applications in practical scenarios are hindered by slow inference speed. Drawing inspiration from the approximation strategies utilized in consistency models, we propose the Sub-path Linear Approximation Model (SLAM), which accelerates diffusion models while maintaining high-quality image generation. SLAM treats the PF-ODE trajectory as a series of PF-ODE sub-paths divided by sampled points, and harnesses sub-path linear (SL) ODEs to form a progressive and continuous error estimation along %the endpoints of each individual PF-ODE sub-path. The optimization on such SL-ODEs allows SLAM to construct denoising mappings with smaller cumulative approximated errors. An efficient distillation method is also developed to facilitate the incorporation of more advanced diffusion models, such as latent diffusion models. Our extensive experimental results demonstrate that SLAM achieves an efficient training regimen, requiring only 6 A100 GPU days to produce a high-quality generative model capable of 2 to 4-step generation with high performance. Comprehensive evaluations on LAION, MS COCO 2014, and MS COCO 2017 datasets also illustrate that SLAM surpasses existing acceleration methods in few-step generation tasks, achieving state-of-the-art performance both on FID and the quality of the generated images.

While recent advances in 3D-aware Generative Adversarial Networks (GANs) have aided the development of near-frontal view human face synthesis, the challenge of comprehensively synthesizing a full 3D head viewable from all angles still persists. Although PanoHead proves the possibilities of using a large-scale dataset with images of both frontal and back views for full-head synthesis, it often causes artifacts for back views. Based on our in-depth analysis, we found the reasons are mainly twofold. First, from network architecture perspective, we found each plane in the utilized tri-plane/tri-grid representation space tends to confuse the features from both sides, causing mirroring'' artifacts (e.g., the glasses appear in the back). Second, from data supervision aspect, we found that existing discriminator training in 3D GANs only focuses on the quality of the rendered image itself, and does not care about its plausibility with the perspective from which it was rendered. This makes it possible to generateface'' in the non-frontal view, due to its easiness to fool the discriminator. In response, we propose SphereHead, a novel tri-plane representation in the spherical coordinate system that fits the human head's geometric characteristics and efficiently mitigates many of the generated artifacts. We further introduce a view-image consistency …

Generating realistic audio for human interactions is important for many applications, such as creating sound effects for films or virtual reality games. Existing approaches implicitly assume total correspondence between the video and audio during training, yet many sounds happen off-screen and have weak to no correspondence with the visuals---resulting in uncontrolled ambient sounds or hallucinations at test time. We propose a novel ambient-aware audio generation model, AV-LDM. We devise a novel audio-conditioning mechanism to learn to disentangle foreground action sounds from the ambient background sounds in in-the-wild training videos. Given a novel silent video, our model uses retrieval-augmented generation to create audio that matches the visual content both semantically and temporally. We train and evaluate our model on two in-the-wild egocentric video datasets Ego4D and EPIC-KITCHENS. Our model outperforms an array of existing methods, allows controllable generation of the ambient sound, and even shows promise for generalizing to computer graphics game clips. Overall, our work is the first to focus video-to-audio generation faithfully on the observed visual content despite training from uncurated clips with natural background sounds.

In this paper, we introduce a Multimodal Large Language Model-based Generation Assistant (LLMGA), leveraging the vast reservoir of knowledge and proficiency in reasoning, comprehension, and response inherent in Large Language Models (LLMs) to assist users in image generation and editing. Diverging from existing approaches where Multimodal Large Language Models (MLLMs) generate fixed-size embeddings to control Stable Diffusion (SD), our LLMGA provides a detailed language generation prompt for precise control over SD. This not only augments LLM context understanding but also reduces noise in generation prompts, yields images with more intricate and precise content, and elevates the interpretability of the network. To this end, we curate a comprehensive dataset comprising prompt refinement, similar image generation, inpainting \& outpainting, and instruction-based editing. Moreover, we propose a two-stage training scheme. In the first stage, we train the MLLM to grasp the properties of image generation and editing, enabling it to generate detailed prompts. In the second stage, we optimize SD to align with the MLLM's generation prompts. Additionally, we propose a reference-based restoration network to alleviate texture, brightness, and contrast disparities between generated and preserved regions during inpainting and outpainting. Extensive results show that LLMGA has promising generation and editing capabilities and can …

Manually creating textures for 3D meshes is time-consuming, even for expert visual content creators. We propose a fast approach for automatically texturing an input 3D mesh based on a user-provided text prompt. Importantly, our approach disentangles lighting from surface material/reflectance in the resulting texture so that the mesh can be properly relit and rendered in any lighting environment. Our method introduces LightControlNet, a new text-to-image model based on the ControlNet architecture, that allows the specification of the desired lighting as a conditioning image to the model. Our text-to-texture pipeline then constructs the texture in two stages. The first stage produces a sparse set of visually consistent reference views of the mesh using LightControlNet. The second stage applies a texture optimization based on Score Distillation Sampling (SDS) that works with LightControlNet to increase the texture quality while disentangling surface material from lighting. We show that this pipeline is significantly faster than previous text-to-texture methods, while producing high-quality and relightable textures.

Creating photorealistic avatars for individuals traditionally involves extensive capture sessions with complex and expensive studio devices like the LightStage system. While recent strides in neural representations have enabled the generation of photorealistic and animatable 3D avatars from quick phone scans, they have baked-in capture-time lighting, lack facial details and have missing regions in areas such as the back of the ears. Thus, they lag in quality compared to studio-captured avatars. In this paper, we propose a method to bridge this gap by generating studio-like illuminated texture maps from short, monocular phone captures. We do this by parameterizing the phone texture maps using the W+ space of a StyleGAN2, enabling near-perfect reconstruction. Then, we finetune a StyleGAN2 by sampling in the W+ parameterized space using a very small set of studio-captured textures as an adversarial training signal. To further enhance the realism and accuracy of facial details, we super-resolve the output of the StyleGAN2 using a diffusion model conditioned on image gradients of the phone-captured texture map. Once trained, our method excels at producing studio-like facial texture maps from casual monocular smartphone videos. Demonstrating its capabilities, we showcase the generation of photorealistic, uniformly lit, complete avatars from monocular phone captures.

Texturing 3D humans with semantic UV maps remains a challenge due to the difficulty of acquiring reasonably unfolded UV. Despite recent text-to-3D advancements in supervising multi-view renderings using large text-to-image (T2I) models, issues persist with generation speed, text consistency, and texture quality, resulting in data scarcity among existing datasets. We present TexDreamer, the first zero-shot multimodal high-fidelity 3D human texture generation model. Utilizing an efficient texture adaptation finetuning strategy, we adapt large T2I model to a semantic UV structure while preserving its original generalization capability. Leveraging a novel feature translator module, the trained model is capable of generating high-fidelity 3D human textures from either text or image within seconds. Furthermore, we introduce ArTicuLated humAn textureS (ATLAS), the largest high-resolution (1, 024 × 1, 024) 3D human texture dataset which contains 50k high-fidelity textures with text descriptions. Our dataset and model will be available for research purposes.

In this work, we tackle the challenge of enhancing the realism and expressiveness in talking head video generation by focusing on the dynamic and nuanced relationship between audio cues and facial movements. We identify the limitations of traditional techniques that often fail to capture the full spectrum of human expressions and the uniqueness of individual facial styles. To address these issues, we propose EMO, a novel framework that utilizes a direct audio-to-video synthesis approach, bypassing the need for intermediate 3D models or facial landmarks. Our method ensures seamless frame transitions and consistent identity preservation throughout the video, resulting in highly expressive and lifelike animations. Experimental results demonsrate that EMO is able to produce not only convincing speaking videos but also singing videos in various styles, significantly outperforming existing state-of-the-art methodologies in terms of expressiveness and realism.

We present a novel approach for synthesizing emotion-controllable 3D talking heads, featuring enhanced lip synchronization and rendering quality. Despite significant progress in the field, prior methods still suffer from multi-view consistency and a lack of emotional expressiveness. To address these issues, we collect EmoTalk3D dataset with calibrated multi-view videos, emotional annotations, and per-frame 3D geometry. By training on the EmoTalk3D dataset, we propose a 'Speech-to-Geometry-to-Appearance' mapping framework that first predicts faithful 3D geometry sequence from the audio features, then the appearance of a 3D talking head represented by 4D Gaussians is synthesized from the predicted geometry. The appearance is further disentangled into canonical and dynamic Gaussians, learned from multi-view videos, and fused to render free-view talking head animation. Moreover, our model extracts emotion labels from the input speech and enables controllable emotion in the generated talking heads. Our method exhibits improved rendering quality and stability in lip motion generation while capturing dynamic facial details such as wrinkles and subtle expressions. Experiments demonstrate the effectiveness of our approach in generating high-fidelity and emotion-controllable 3D talking heads. The code and EmoTalk3D dataset will be publicly released upon publication.

Creating high-fidelity 3D human head avatars is crucial for applications in VR/AR, telepresence, digital human interfaces, and film production. Recent advances have leveraged morphable face models to generate animated head avatars from easily accessible data, representing varying identities and expressions within a low-dimensional parametric space. However, existing methods often struggle with modeling complex appearance details, e.g., hairstyles and accessories, and suffer from low rendering quality and efficiency. This paper introduces a novel approach, 3D Gaussian Parametric Head Model, which employs 3D Gaussians to accurately represent the complexities of the human head, allowing precise control over both identity and expression. Additionally, it enables seamless face portrait interpolation and the reconstruction of detailed head avatars from a single image. Unlike previous methods, the Gaussian model can handle intricate details, enabling realistic representations of varying appearances and complex expressions. Additionally, this paper presents a well-designed training framework to ensure smooth convergence, providing a robust guarantee for learning the rich content. Our method achieves high-quality, photo-realistic rendering with real-time efficiency, making it a valuable contribution to the field of parametric head models.

Modern avatar generators allow anyone to synthesize photorealistic real-time talking avatars, ushering in a new era of avatar-based human communication, such as with immersive AR/VR interactions or videoconferencing with limited bandwidths. Their safe adoption, however, requires a mechanism to verify if the rendered avatar is trustworthy: does it use the appearance of an individual without their consent? We term this task avatar fingerprinting. To tackle it, we first introduce a large-scale dataset of real and synthetic videos of people interacting on a video call, where the synthetic videos are generated using the facial appearance of one person and the expressions of another. We verify the identity driving the expressions in a synthetic video, by learning motion signatures that are independent of the facial appearance shown. Our solution, the first in this space, achieves an average AUC of 0.85. Critical to its practical use, it also generalizes to new generators never seen in training (average AUC of 0.83).

Modeling and rendering photorealistic avatars is of crucial importance in many applications. Existing methods that build a 3D avatar from visual observations, however, struggle to reconstruct clothed humans. We introduce PhysAvatar, a novel framework that combines inverse rendering with inverse physics to automatically estimate the shape and appearance of a human from multi-view video data along with the physical parameters of the fabric of their clothes. For this purpose, we adopt a mesh-aligned 4D Gaussian technique for spatio-temporal mesh tracking as well as a physically based inverse renderer to estimate the intrinsic material properties. PhysAvatar integrates a physics simulator to estimate the physical parameters of the garments using gradient-based optimization in a principled manner. These novel capabilities enable PhysAvatar to create high-quality novel-view renderings of avatars dressed in loose-fitting clothes under motions and lighting conditions not seen in the training data. This marks a significant advancement towards modeling photorealistic digital humans using physically based inverse rendering with physics in the loop.

Existing portrait relighting methods struggle with precise control over facial shadows, particularly when faced with challenges such as handling hard shadows from directional light sources or adjusting shadows while remaining in harmony with existing lighting conditions. In many situations, completely altering input lighting is undesirable for portrait retouching applications: one may want to preserve some authenticity in the captured environment. Existing shadow editing methods typically restrict their application to just the facial region and often offer limited lighting control options, such as shadow softening or rotation. In this paper, we introduce COMPOSE: a novel shadow editing pipeline for human portraits, offering precise control over shadow attributes such as shape, intensity, and position, all while preserving the original environmental illumination of the portrait. This level of disentanglement and controllability is obtained thanks to a novel decomposition of the environment map representation into ambient light and an editable gaussian dominant light source. COMPOSE is a four-stage pipeline that consists of light estimation and editing, light diffusion, shadow synthesis, and finally shadow editing. We define facial shadows as the result of a dominant light source, encoded using our novel gaussian environment map representation. Utilizing an OLAT dataset, we have trained models to: (1) …

The majority of existing Low-light Image Enhancement (LLIE) methods attempt to directly learn the mapping from Low-Light (LL) to Normal-Light (NL) images or benefit from semantic or illumination map to guide such learning. However, the inherent ill-posed nature of LLIE, coupled with the additional difficulty in retrieving the semantic information from significantly impaired inputs, compromises the performance of enhanced outputs, especially in extremely low-light environments. To address this issue, we present a new LLIE network via Generative LAtent feature based codebook REtrieval (GLARE) to improve the visibility of LL images. The codebook prior is derived from undegraded NL images using Vector Quantization (VQ) strategy. However, simply adopting codebook prior does not necessarily ensure the alignment between LL and NL features. Therefore, we develop an Invertible Latent Normalizing Flow (I-LNF) to generate features aligned with NL latent representations, guaranteeing the correct code matching in the codebook. In addition, a novel Adaptive Feature Transformation (AFT) module, which contains an Adaptive Mix-up Block (AMB) and a dual-decoder architecture, is devised to further elevate the fidelity while maintaining realistic details provided by codebook prior. Extensive experiments verify the superior performance of our GLARE. More importantly, the application on low-light object detection demonstrates the effectiveness …

High dynamic range (HDR) imaging involves capturing a series of frames of the same scene, each with different exposure settings, to broaden the dynamic range of light. This can be achieved through burst capturing or using staggered HDR sensors that capture long and short exposures simultaneously in the camera image signal processor (ISP). Within camera ISP pipeline, illuminant estimation is a crucial step aiming to estimate the color of the global illuminant in the scene. This estimation is used in camera ISP white-balance module to remove undesirable color cast in the final image. Despite the multiple frames captured in the HDR pipeline, conventional illuminant estimation methods often rely only on a single frame of the scene. In this paper, we explore leveraging information from frames captured with different exposure times. Specifically, we introduce a simple feature extracted from dual-exposure images to guide illuminant estimators, referred to as the dual-exposure feature (DEF). To validate the efficiency of DEF, we employed two illuminant estimators using the proposed DEF: 1) a multilayer perceptron network (MLP), referred to as exposure-based MLP (EMLP), and 2) a modified version of the convolutional color constancy (CCC) to integrate our DEF, that we call ECCC. Both EMLP and …

Typical projectors are designed to programmably display 2D content at a single depth. In this work, we explore how to engineer a depth-varying projector system that is capable of forming desired patterns at multiple depths. To this end, we leverage a holographic approach, but a naive implementation of such a system is limited in its depth programmability. Inspired by recent work in near-eye displays, we add a lens array to a holographic projector to maximize the depth variation of the projected content, for which we propose an optimization-driven calibration method. We demonstrate a number of applications using this system, including novel 3D interfaces for future wearables, privacy-preserving projection, depth sensing, and light curtains.

Neural implicit representation of visual scenes has attracted a lot of attention in recent research of computer vision and graphics. Most prior methods focus on how to reconstruct 3D scene representation from a set of images. In this work, we demonstrate the possibility to recover the neural radiance fields (NeRF) from a single blurry image and its corresponding event stream. We model the camera motion with a cubic B-Spline in SE(3) space. Both the blurry image and the brightness change within a time interval, can then be synthesized from the 3D scene representation given the 6-DoF poses interpolated from the cubic B-Spline. Our method can jointly learn both the implicit neural scene representation and recover the camera motion by minimizing the differences between the synthesized data and the real measurements without pre-computed camera poses from COLMAP. We evaluate the proposed method with both synthetic and real datasets. The experimental results demonstrate that we are able to render view-consistent latent sharp images from the learned NeRF and bring a blurry image alive in high quality.

Neural rendering-based urban scene reconstruction methods commonly rely on images collected from driving vehicles with cameras facing and moving forward. Although these methods can successfully synthesize from views similar to training camera trajectory, directing the novel view outside the training camera distribution does not guarantee on-par performance. In this paper, we tackle the Extrapolated View Synthesis (EVS) problem by evaluating the reconstructions on views such as looking left, right or downwards with respect to training camera distributions. To improve rendering quality for EVS, we initialize our model by constructing dense LiDAR map, and propose to leverage prior scene knowledge such as surface normal estimator and large-scale diffusion model. Qualitative and quantitative comparisons demonstrate the effectiveness of our methods on EVS. To the best of our knowledge, we are the first to address the EVS problem in urban scene reconstruction. We will release the code upon acceptance.

Large scale 3D scene reconstruction is important for applications such as virtual reality and simulation. Existing neural rendering approaches (e.g., NeRF, 3DGS) have achieved realistic reconstructions on large scenes, but optimize per scene, which is expensive and slow, and exhibit noticeable artifacts under large view changes due to overfitting. Generalizable approaches are fast, but primarily work for small scenes/objects and often produce lower quality rendering results. In this work, we introduce G3R, a generalizable reconstruction approach that can efficiently predict high-quality 3D scene representations for large scenes. We propose to learn a reconstruction network that takes the gradient feedback signals from differentiable rendering to iteratively update a 3D scene representation, combining the benefits of high photorealism from per-scene optimization with data-driven priors from fast feed-forward prediction methods. Experiments on large-scale urban-driving and drone datasets show that G3R accelerates the reconstruction process by at least 10x while achieving comparable or better realism compared to 3DGS, and also being more robust to large view changes.

We propose an application of online hard sample mining for efficient training of Neural Radiance Fields (NeRF). NeRF models produce state-of-the-art quality for many 3D reconstruction and rendering tasks but require substantial computational resources. The encoding of the scene information within the NeRF network parameters necessitates stochastic sampling. We observe that during the training, a major part of the compute time and memory usage is spent on processing already learnt samples, which no longer affect the model update significantly. We identify the backward pass on the stochastic samples as the computational bottleneck during the optimization. We thus perform the first forward pass in inference mode as a relatively low-cost search for hard samples. This is followed by building the computational graph and updating the NeRF network parameters using only the hard samples. To demonstrate the effectiveness of the proposed approach, we apply our method to Instant-NGP, resulting in significant improvements of the view-synthesis quality over the baseline (1 dB improvement on average per training time, or 2x speedup to reach the same PSNR level) along with 40% memory savings coming from using only the hard samples to build the computational graph. As our method only interfaces with the network module, …

Recent efforts in using 3D Gaussians for scene reconstruction and novel view synthesis can achieve impressive results on curated benchmarks; however, images captured in real life are often blurry. In this work, we analyze the robustness of Gaussian-Splatting-based methods against various image blur, such as motion blur, defocus blur, downscaling blur, etc. Under these degradations, Gaussian-Splatting-based methods tend to overfit and produce worse results than Neural-Radiance-Field-based methods. To address this issue, we propose Blur Agnostic Gaussian Splatting (BAGS). BAGS introduces additional 2D modeling capacities such that a 3D-consistent and high quality scene can be reconstructed despite image-wise blur. Specifically, we model blur by estimating per-pixel convolution kernels from a Blur Proposal Network (BPN). BPN is designed to consider spatial, color, and depth variations of the scene to maximize modeling capacity. Additionally, BPN also proposes a quality-assessing mask, which indicates regions where blur occur. Finally, we introduce a coarse-to-fine kernel optimization scheme; this optimization scheme is fast and avoids sub-optimal solutions due to a sparse point cloud initialization, which often occurs when we apply Structure-from-Motion on blurry images. We demonstrate that BAGS achieves photorealistic renderings under various challenging blur conditions and imaging geometry, while significantly improving upon existing approaches.

The ability to distill object-centric abstractions from intricate visual scenes underpins human-level generalization. Despite the significant progress in object-centric learning methods, learning object-centric representations in the 3D physical world remains a crucial challenge. In this work, we propose SlotLifter, a novel object-centric radiance model that aims to address the challenges of scene reconstruction and decomposition via slot-guided feature lifting. Such a design unites object-centric learning representations and image-based rendering methods, offering state-of-the-art performance in scene decomposition and novel-view synthesis on four challenging synthetic and four complex real-world datasets, outperforming existing 3D object-centric learning methods by a large margin. Through extensive ablative studies, we showcase the efficacy of each design in SlotLifter, shedding light on key insights for potential future directions.

Neural Radiance Fields (NeRF) have become increasingly popular for their ability to reconstruct 3D scenes and create new viewpoints with outstanding quality. However, their effectiveness is hindered by rolling shutter (RS) effects commonly found in most camera systems. To solve this, we present RS-NeRF, a method designed to synthesize normal images from novel views using input with RS distortions. This involves a physical model that replicates the image formation process under RS conditions and jointly optimizes NeRF parameters and camera extrinsic for each image row. We further address the inherent shortcomings of the basic RS-NeRF model by delving into RS characteristics and developing algorithms to enhance its functionality. First, we impose a smoothness regularization to better estimate trajectories and improve synthesis quality, in line with the camera movement prior. We also identify and address a fundamental flaw in the vanilla RS model by introducing a multi-sampling algorithm. This new approach greatly improves the model's performance by comprehensively exploiting the RGB data across different rows for each intermediate camera pose. Through rigorous experimentation, we demonstrate that RS-NeRF surpasses previous methods in both synthetic and real-world scenarios, proving its ability to correct RS-related distortions effectively.

Spherical or omni-directional images offer an immersive visual format appealing to a wide range of computer vision applications. However, geometric properties of spherical images pose a major challenge for models and metrics designed for ordinary 2D images. For image generation, this poses a full-stack problem: from smaller data-sets to access to tools that can evaluate quality of models. Specifically, we show that direct application of the established evaluation metric Fréchet Inception Distance (FID) is insufficient for quantifying geometric fidelity in spherical images. We introduce two quantitative metrics accounting for geometric constraints, namely Omnidirectional FID (OmniFID) and Discontinuity Score (DS). OmniFID is an extension of FID, tailored to additionally capture field-of-view requirements of the spherical format by leveraging cubemap projections. DS is a kernel-based seam alignment score of continuity across borders of 2D representations of spherical images. In experiments, OmniFID and DS quantify geometry fidelity issues that are undetected by FID.

Differentiable surface rendering has significantly advanced 3D reconstruction. Existing surface rendering methods assume that the local surface is planar, and thus employ linear approximation based on the Singed Distance Field (SDF) values to predict the point on the surface. However, this assumption overlooks the inherently irregular and non-planar nature of object surfaces in the real world. Consequently, the approximate points tend to deviate from the zero-level set, affecting the fidelity of the reconstructed shape. In this paper, we propose a novel surface rendering method termed CPT-VR, which leverages the Closet Point Transform (CPT) and View and Reflection direction vectors to enhance the quality of reconstruction. Specifically, leveraging the physical property of CPT that accurately projects points near the surface onto the zero-level set, we correct the deviated points, thus achieving an accurate geometry representation. Based on our accurate geometry representation, incorporating the reflection vector into our method can facilitate the appearance modeling of specular regions. Moreover, to enable our method to no longer be dependent on any prior knowledge of the background, we present a background model to learn the background appearance. Compared to previous state-of-the-art methods, CPT-VR achieves better surface reconstruction quality, even for cases with complex structures and …

Faithful human performance capture and free-view rendering from sparse RGB observations is a long-standing problem in Vision and Graphics. The main challenges are the lack of observations and the inherent ambiguities of the setting, e.g. occlusions and depth ambiguity. As a result, radiance fields, which have shown great promise in capturing high-frequency appearance and geometry details in dense setups, perform poorly when naively supervising them on sparse camera views, as the field simply overfits to the sparse input views.To address this, we propose MetaCap, a method for efficient and high-quality geometry recovery and novel view synthesis given very sparse or even a single view of the human. Our key idea is to meta-learn radiance field weights solely from multi-view videos, which can serve as a prior when fine-tuning them on sparse imagery depicting the human. Due to the articulated structure of the human body and motion-induced surface deformations, learning such a prior is non-trivial. Therefore, we propose to meta-learn the field weights in a pose canonical space, which reduces the spatial feature range and makes feature learning more effective. Consequently, one can fine-tune our field parameters to quickly generalize to unseen poses, novel illumination conditions and novel and sparse (even …

X-ray is widely applied for transmission imaging due to its stronger penetration than natural light. When rendering novel view X-ray projections, existing methods mainly based on NeRF suffer from long training time and slow inference speed. In this paper, we propose a 3D Gaussian splatting-based framework, namely X-Gaussian, for X-ray novel view synthesis. Firstly, we redesign a radiative Gaussian point cloud model inspired by the isotropic nature of X-ray imaging. Our model excludes the influence of view direction when learning to predict the radiation intensity of 3D points. Based on this model, we develop a Differentiable Radiative Rasterization (DRR) with CUDA implementation. Secondly, we customize an Angle-pose Cuboid Uniform Initialization (ACUI) strategy that directly uses the parameters of the X-ray scanner to compute the camera information and then uniformly samples point positions within a cuboid enclosing the scanned object. Experiments show that our X-Gaussian outperforms state-of-the-art methods by 6.5 dB while enjoying less than 15% training time and over 73x inference speed. The application on sparse-view CT reconstruction also reveals the practical values of our method. Code and models will be released to the public.

Advances in rendering have led to tremendous growth in texture assets, including resolution, complexity, and novel textures components, but this growth in data volume has not been matched by advances in its compression. Meanwhile Neural Image Compression (NIC) has advanced significantly and shown promising results, but the proposed methods cannot be directly adapted to neural texture compression. First, texture compression requires on-demand and real-time decoding with random access during parallel rendering (e.g. block texture decompression on GPUs). Additionally, NIC does not support multi-resolution reconstruction (mip-levels), nor does it have the ability to efficiently jointly compress different sets of texture channels. In this work, we introduce a novel approach to texture set compression that integrates traditional GPU texture representation and NIC techniques, designed to enable random access and support many-channel texture sets. To achieve this goal, we propose an asymmetric auto-encoder framework that employs a convolutional encoder to capture detailed information in a bottleneck-latent space, and at decoder side we utilize a fully connected network, whose inputs are sampled latent features plus positional information, for a given texture coordinate and mip level. This latent data is defined to enable simplified access to multi-resolution data by simply changing the scanning strides. Experimental …

Recently, 3D Gaussian Splatting (3DGS) has emerged as an efficient approach for accurately representing scenes. However, despite its superior novel view synthesis capabilities, extracting the geometry of the scene directly from the Gaussian properties remains a challenge, as those are optimized based on a photometric loss. While some concurrent models have tried adding geometric constraints during the Gaussian optimization process, they still produce noisy, unrealistic surfaces. We propose a novel approach for bridging the gap between the noisy 3DGS representation and the smooth 3D mesh representation, by injecting real-world knowledge into the depth extraction process. Instead of extracting the geometry of the scene directly from the Gaussian properties, we instead extract the geometry through a pre-trained stereo-matching model. We render stereo-aligned pairs of images corresponding to the original training poses, feed the pairs into a stereo model to get a depth profile, and finally fuse all of the profiles together to get a single mesh. The resulting reconstruction is smoother, more accurate and shows more intricate details compared to other methods for surface reconstruction from Gaussian Splatting, while only requiring a small overhead on top of the fairly short 3DGS optimization process. We performed extensive testing of the proposed method …

3D Gaussian Splatting (3DGS) has shown remarkable success in synthesizing novel views given multiple views of a static scene. Yet, 3DGS faces challenges when applied to dynamic scenes because 3D Gaussian parameters need to be updated per timestep, requiring a large amount of memory and at least a dozen observations per timestep. To address these limitations, we present a compact dynamic 3D Gaussian representation that models positions and rotations as functions of time with a few parameter approximations while keeping other properties of 3DGS including scale, color and opacity invariant. Our method can dramatically reduce memory usage and relax a strict multi-view assumption. In our experiments on monocular and multi-view scenarios, we show that our method not only matches state-of-the-art methods, often linked with slower rendering speeds, in terms of high rendering quality but also significantly surpasses them by achieving a rendering speed of 118 frames per second (FPS) at a resolution of 1,352x1,014 on a single GPU.

Interactive segmentation of 3D Gaussians opens a great opportunity for real-time manipulation of 3D scenes thanks to the real-time rendering capability of 3D Gaussian Splatting. However, the current methods suffer from time-consuming post-processing to deal with noisy segmentation output. Also, they struggle to provide detailed segmentation, which is important for fine-granular manipulation of 3D scenes. In this study, we propose Click-Gaussian, which learns distinguishable feature fields of two-level granularity, facilitating segmentation without time-consuming post-processing.We delve into challenges stemming from inconsistently learned feature fields resulting from 2D segmentation obtained independently from a 3D scene. 3D segmentation accuracy deteriorates when 2D segmentation results across the views, primary cues for 3D segmentation, are in conflict. To overcome these issues, we propose Global Feature-guided Learning (GFL). GFL constructs the clusters of global feature candidates from noisy 2D segments across the views, which smooths out noises when learning the features of 3D Gaussians. Our method runs in 10ms per click, 15 to 130 times as fast as the previous methods, while also significantly improving segmentation accuracy.

Iso-surface extraction from an implicit field is a fundamental process in various applications of computer vision and graphics. When dealing with geometric shapes with complicated geometric details, many existing algorithms suffer from high computational costs and memory usage. This paper proposes McGrids, a novel approach to improve the efficiency of iso-surface extraction. The key idea is to construct adaptive grids for iso-surface extraction rather than using a simple uniform grid as prior art does. Specifically, we formulate the problem of constructing adaptive grids as a probability sampling problem, which is then solved by Monte Carlo process. We demonstrate McGrids' capability with extensive experiments from both analytical SDFs computed from surface meshes and learned implicit fields from real multiview images. The experiment results show that our McGrids can significantly reduce the number of implicit field queries, resulting in significant memory reduction, while producing high-quality meshes with rich geometric details.

We present latentSplat, a method to predict variational Gaussians in a 3D latent space that can be splatted and decoded by a light-weight generative 2D architecture. Existing methods for generalizable 3D reconstruction either do not enable fast inference of high resolution novel views due to slow volume rendering, or are limited to interpolation of close input views, even in simpler settings with a single central object, where 360-degree generalization is possible. In this work, we combine a regression-based approach with a generative model, moving towards both of these capabilities within the same method, trained purely on readily available real video data. The core of our method are variational 3D Gaussians, a representation that efficiently encodes varying uncertainty within a latent space consisting of 3D feature Gaussians. From these Gaussians, specific instances can be sampled and rendered via efficient Gaussian splatting and a fast, generative decoder network. We show that latentSplat outperforms previous works in reconstruction quality and generalization, while being fast and scalable to high-resolution data.

We introduce a novel non-parametric sensor noise model that directly constructs probability mass functions per intensity level from captured images. We show that our noise model provides a more accurate fit to real sensor noise than existing models. We detail the capture procedure for deriving our non-parametric noise model and introduce an interpolation method that reduces the number of ISOs levels that need to be captured. In addition, we propose a method to synthesize noise on existing noisy images when noise-free images are not available. Our noise model is straightforward to calibrate and provides notable improvements over competing noise models on downstream tasks.

We propose UpFusion, a system that can perform novel view synthesis and infer 3D representations for an object given a sparse set of reference images without corresponding pose information. Current sparse-view 3D inference methods typically rely on camera poses to geometrically aggregate information from input views, but are not robust in-the-wild when such information is unavailable/inaccurate. In contrast, UpFusion sidesteps this requirement by learning to implicitly leverage the available images as context in a conditional generative model for synthesizing novel views. We incorporate two complementary forms of conditioning into diffusion models for leveraging the input views: a) via inferring query-view aligned features using a scene-level transformer, b) via intermediate attentional layers that can directly observe the input image tokens. We show that this mechanism allows generating high-fidelity novel views while improving the synthesis quality given additional (unposed) images. We evaluate our approach on the Co3Dv2 and Google Scanned Objects datasets and demonstrate the benefits of our method over pose-reliant sparse-view methods as well as single-view methods that cannot leverage additional views. Finally, we also show that our learned model can generalize beyond the training categories and even allow reconstruction from self-captured images of generic objects in-the-wild.

Denoising diffusion models have demonstrated outstanding results in 2D image generation, yet it remains a challenge to replicate its success in 3D shape generation. In this paper, we propose leveraging multi-view depth, which represents complex 3D shapes in a 2D data format that is easy to denoise. We pair this representation with a diffusion model, MVDD, that is capable of generating high-quality dense point clouds with 20K+ points with fine-grained details. To enforce 3D consistency in multi-view depth, we introduce an epipolar line segment attention that conditions the denoising step for a view on its neighboring views. Additionally, a depth fusion module is incorporated into diffusion steps to further ensure the alignment of depth maps. When augmented with surface reconstruction, MVDD can also produce high-quality 3D meshes. Furthermore, MVDD stands out in other tasks such as depth completion, and can serve as a 3D prior, significantly boosting many downstream tasks, such as GAN inversion. State-of-the-art results from extensive experiments demonstrate MVDD's excellent ability in 3D shape generation, depth completion, and its potential as a 3D prior for downstream tasks.

3D content creation has achieved significant progress in terms of both quality and speed. Although current feed-forward models can produce 3D objects in seconds, their resolution is constrained by the intensive computation required during training. In this paper, we introduce Large Multi-View Gaussian Model (LGM), a novel framework designed to generate high-resolution 3D models from text prompts or single-view images. Our key insights are two-fold: 1) 3D Representation: We propose multi-view Gaussian features as an efficient yet powerful representation, which can then be fused together for differentiable rendering. 2) 3D Backbone: We present an asymmetric U-Net as a high-throughput backbone operating on multi-view images, which can be produced from text or single-view image input by leveraging multi-view diffusion models. Extensive experiments demonstrate the high fidelity and efficiency of our approach. Notably, we maintain the fast speed to generate 3D objects within 5 seconds while boosting the training resolution to 512, thereby achieving high-resolution 3D content generation.

In this paper, we introduce a technique to estimate measured BRDFs from a sparse set of samples. Our approach offers accurate BRDF reconstructions that are generalizable to new materials. This opens the door to BDRF reconstructions from a variety of data sources. The success of our approach relies on the ability of hypernetworks to generate a robust representation of BRDFs and a set encoder that allows us to feed inputs of different sizes to the architecture. The set encoder and the hypernetwork also enable the compression of densely sampled BRDFs. We evaluate our technique both qualitatively and quantitatively on the well-known MERL dataset of 100 isotropic materials. Our approach accurately 1) estimates the BRDFs of unseen materials even for an extremely sparse sampling, 2) compresses the measured BRDFs into very small embeddings, e.g., 7D.

3D vision is inherently characterized by sparse spatial structures, which propels the necessity for an efficient paradigm tailored to 3D generation. Another discrepancy is the amount of training data, which undeniably affects generalization if we only use limited 3D data. To solve these, we design a 3D generation framework that maintains most of the building blocks of StableDiffusion with minimal adaptations for textured shape generation. We design a Sparse Encoding Module for details preservation and an Adversarial Decoding Module for better shape recovery. Moreover, we clean up data and build a benchmark on the biggest 3D dataset (Objaverse). We drop the concept of `specific class' and treat the 3D Textured Shapes Generation as an open-vocabulary problem. We first validate our network design on ShapeNetV2 with 55K samples on single-class unconditional generation and multi-class conditional generation tasks. Then we report metrics on processed Objaverse-clean with 200K samples on the image conditional generation task. Extensive experiments demonstrate our proposal outperforms SOTA methods and takes a further step towards open-vocabulary 3D generation.

We present Structured Neural Radiance Field (Structured-NeRF) for indoor scene representaion based on a novel hierarchical scene graph structure to organize the neural radiance field. Existing object-centric methods focus only on the inherent characteristics of objects, while overlooking the semantic and physical relationships between them. Our scene graph is adept at managing the complex real-world correlation between objects within a scene, enabling functionality beyond novel view synthesis, such as scene re-arrangement. Based on the hierarchical structure, we introduce the optimization strategy based on semantic and physical relationships, thus simplifying the operations involved in scene editing and ensuring both efficiency and accuracy. Moreover, we conduct shadow rendering on objects to further intensify the realism of the rendered images. Experimental results demonstrate our structured representation not only achieves state-of-the-art (SOTA) performance in object-level and scene-level rendering, but also advances downstream applications in union with LLM/VLM, such as automatic and instruction/image conditioned scene re-arrangement, thereby extending the NeRF to interactive editing conveniently and controllably.

The current GAN inversion methods typically can only edit the appearance and shape of a single object and background while overlooking spatial information. In this work, we propose a 3D editing framework, 3D-GOI, to enable multifaceted editing of affine information (scale, translation, and rotation) on multiple objects. 3D-GOI realizes the complex editing function by inverting the abundance of attribute codes (object shape/appearance/scale/rotation/translation, background shape/appearance, and camera pose) controlled by GIRAFFE, a renowned 3D GAN. Accurately inverting all the codes is challenging, 3D-GOI solves this challenge following three main steps. First, we segment the objects and the background in a multi-object image. Second, we use a custom Neural Inversion Encoder to obtain coarse codes of each object. Finally, we use a round-robin optimization algorithm to get precise codes to reconstruct the image. To the best of our knowledge, 3D-GOI is the first framework to enable multifaceted editing on multiple objects. Both qualitative and quantitative experiments demonstrate that 3D-GOI holds immense potential for flexible, multifaceted editing in complex multi-object scenes.Our project code will be made publicly available.

3D Gaussian splatting, emerging as a groundbreaking approach, has drawn increasing attention for its capabilities of high-fidelity reconstruction and real-time rendering. However, it couples the appearance and geometry of the scene within the Gaussian attributes, which hinders the flexibility of editing operations, such as texture swapping. To address this issue, we propose a novel approach, namely Texture-GS, to disentangle the appearance from the geometry by representing it as a 2D texture, thereby facilitating appearance editing. Technically, the disentanglement is achieved by our proposed texture mapping module, which consists of a UV mapping MLP to learn the UV coordinates for the 3D Gaussian centers, a local Taylor expansion of the MLP to efficiently approximate the UV coordinates for the ray-Gaussian intersections, and a learnable texture to capture the fine-grained appearance. Extensive experiments on the DTU dataset demonstrate that our method not only facilitates high-fidelity appearance editing but also achieves real-time rendering on consumer-level devices, e.g. a single RTX 2080 Ti GPU.

Recent research on texture synthesis for 3D shapes benefits a lot from dramatically developed 2D text-to-image diffusion models, including inpainting-based and optimization-based approaches. However, these methods ignore the modal gap between the 2D diffusion model and 3D objects, which primarily render 3D objects into 2D images and texture each image separately. In this paper, we revisit the texture synthesis and propose a Variance alignment based 3D-2D Collaborative Denoising framework, dubbed VCD-Texture, to address these issues. Formally, we first unify both 2D and 3D latent feature learning in diffusion self-attention modules with re-projected 3D attention receptive fields. Subsequently, the denoised multi-view 2D latent features are aggregated into 3D space and then rasterized back to formulate more consistent 2D predictions. However, the rasterization process suffers from an intractable variance bias, which is theoretically addressed by the proposed variance alignment, achieving high-fidelity texture synthesis. Moreover, we present an inpainting refinement to further improve the details with conflicting regions. Notably, there is not a publicly available benchmark to evaluate texture synthesis, which hinders its development. Thus we construct a new evaluation set built upon three open-source 3D datasets and propose to use four metrics to thoroughly validate the texturing performance. Comprehensive experiments demonstrate that …

Recent advancements in text-to-3D generation technology have significantly advanced the conversion of textual descriptions into imaginative well-geometrical and finely textured 3D objects. Despite these developments, a prevalent limitation arises from the use of RGB data in diffusion or reconstruction models, which often results in models with inherent lighting and shadows effects that detract from their realism, thereby limiting their usability in applications that demand accurate relighting capabilities. To bridge this gap, we present UniDream, a text-to-3D generation framework by incorporating unified diffusion priors. Our approach consists of three main components: (1) a dual-phase training process to get albedo-normal aligned multi-view diffusion and reconstruction models, (2) a progressive generation procedure for geometry and albedo-textures based on Score Distillation Sample (SDS) using the trained reconstruction and diffusion models, and (3) an innovative application of SDS for finalizing PBR generation while keeping a fixed albedo based on Stable Diffusion model. Extensive evaluations demonstrate that UniDream surpasses existing methods in generating 3D objects with clearer albedo textures, smoother surfaces, enhanced realism, and superior relighting capabilities.

By leveraging the text-to-image diffusion prior, score distillation can synthesize 3D contents without paired text-3D training data. Instead of spending hours of online optimization per text prompt, recent studies have been focused on learning a text-to-3D generative network for amortizing multiple text-3D relations, which can synthesize 3D contents in seconds. However, existing score distillation methods are hard to scale up to a large amount of text prompts due to the difficulties in aligning pretrained diffusion prior with the distribution of rendered images from various text prompts. Current state-of-the-arts such as Variational Score Distillation finetune the pretrained diffusion model to minimize the noise prediction error so as to align the distributions, which are however unstable to train and will impair the model's comprehension capability to numerous text prompts. Based on the observation that the diffusion models tend to have lower noise prediction errors at earlier timesteps, we propose Asynchronous Score Distillation (ASD), which minimizes the noise prediction error by shifting the diffusion timestep to earlier ones. ASD is stable to train and can scale up to 100k prompts. It reduces the noise prediction error without changing the weights of pre-trained diffusion model, thus keeping its strong comprehension capability to prompts. We …

Text-to-3D generation, which synthesizes 3D assets according to an overall text description, has significantly progressed. However, a challenge arises when the specific appearances need customizing at designated viewpoints but referring solely to the overall description for generating 3D objects. For instance, ambiguity easily occurs when producing a T-shirt with distinct patterns on its front and back using a single overall text guidance. In this work, we propose DreamView, a text-to-image approach enabling multi-view customization while maintaining overall consistency by adaptively injecting the view-specific and overall text guidance through a collaborative text guidance injection module, which can also be lifted to 3D generation via score distillation sampling. DreamView is trained with large-scale rendered multi-view images and their corresponding view-specific texts to learn to balance the separate content manipulation in each view and the global consistency of the overall object, resulting in a dual achievement of customization and consistency. Consequently, DreamView empowers artists to design 3D objects creatively, fostering the creation of more innovative and diverse 3D assets.

Designing high-quality indoor 3D scenes is important in many practical applications, such as room planning or game development. Conventionally, this has been a time-consuming process which requires both artistic skill and familiarity with professional software, making it hardly accessible for layman users. However, recent advances in generative AI have established solid foundation for democratizing 3D design. In this paper we propose a pioneering approach for text-based 3D room design. Given a prompt in natural language describing the object placement in the room, our method produces a high-quality 3D scene corresponding to it. With an additional text prompt the users can change the appearance of the entire scene or of individual objects in it. Built using in-context learning, CAD model retrieval and 3D-Gaussian-Splatting-based stylization, our turnkey pipeline produces state-of-the-art 3D scenes, while being easy to use even for novices.

This paper addresses the task of 3D clothed human generation from textural descriptions. Previous works usually encode the human body and clothes as a holistic model and generate the whole model in a single-stage optimization, which makes them struggle for clothes editing and meanwhile lose fine-grained control over the whole generation process(e.g., specify the order of inside and outside of clothes). To solve this, we propose a layer-wise clothed human representation combined with a progressive optimization strategy, which produces clothes disentangled 3D human models while providing control capacity for the generation process. The basic idea is progressively generating a minimal-clothed human body and layer-wise clothes. During clothes generation, a novel stratified compositional rendering method is proposed to fuse multi-layer human models, and a new loss function is utilized to help decouple the clothes model from the human body. The proposed method achieves high-quality disentanglement, which thereby provides an effective way for 3D garment generation. Extensive experiments demonstrate that our approach achieves better 3D clothed human generation than the holistic modeling method while also supporting cloth editing applications such as virtual try-on. \keywords{Text-to-3D generation \and Clothed human generation}

With the development of the large-scale diffusion model, Artificial Intelligence Generated Content (AIGC) techniques are popular recently. However, how to truly make it serve our daily lives remains an open question. To this end, in this paper, we focus on employing AIGC techniques in one filed of E-commerce marketing, \ie, generating hyper-realistic advertising images for displaying user-specified shoes by human. Specifically, we propose a shoe-wearing system, called \textbf{ShoeModel}, to generate plausible images of human legs interacting with the given shoes. It consists of three modules: (1) shoe wearable-area detection module (WD), (2) leg-pose synthesis module (LpS) and the final (3) shoe-wearing image generation module (SW). Them three are performed in ordered stages. Compared to baselines, our ShoeModel is shown to generalize better to different type of shoes and has ability of keeping the ID-consistency of the given shoes, as well as automatically producing reasonable interactions with human. Extensive experiments show the effectiveness of our proposed shoe-wearing system. Figure \ref{fig_intro} shows the input and output examples of our ShoeModel.

In this paper, we introduce D4-VTON, a novel solution for image-based virtual try-on that seamlessly replaces a person's original garments with target garments while preserving pose and identity. We address challenges encountered in prior works, such as inaccurate clothing parsers causing artifacts and failing to ensure faithful semantic alignment. Additionally, we tackle the difficulties faced by diffusion models in solving this specific task, which involves the composite tasks of inpainting and denoising. To achieve these goals, we employ two self-contained technologies: Firstly, we propose a Dynamic Group Warping Module (DGWM) to disentangle semantic information and guide warping flows for authentic warped garments. Secondly, we deploy a Differential Noise Restoration Process (DNRP) to capture differential noise between incomplete try-on input and its complete counterpart, facilitating lifelike final results with negligible overhead. Extensive experiments demonstrate that D4-VTON surpasses state-of-the-art methods both quantitatively and qualitatively by a significant margin, showcasing its superiority in generating realistic images and precise semantic alignment.

Neural rendering techniques have significantly advanced 3D human body modeling. However, previous approaches often overlook dynamics induced by factors such as motion inertia, leading to challenges in scenarios like abrupt stops after rotation, where the pose remains static while the appearance changes. This limitation arises from reliance on a single pose as conditional input, resulting in ambiguity in mapping one pose to multiple appearances. In this study, we elucidate that variations in human appearance depend not only on the current frame's pose condition but also on past pose states. Therefore, we introduce Dyco, a novel method utilizing the delta pose sequence representation for non-rigid deformations and canonical space to effectively model temporal appearance variations. To prevent a decrease in the model's generalization ability to novel poses, we further propose low-dimensional global context to reduce unnecessary inter-body part dependencies and a quantization operation to mitigate overfitting of the delta pose sequence by the model. To validate the effectiveness of our approach, we collected a novel dataset named I3D-Human, with a focus on capturing temporal changes in clothing appearance under approximate poses. Through extensive experiments on both I3D-Human and existing datasets, our approach demonstrates superior qualitative and quantitative performance. In addition, our …

We introduce a new method for learning a generative model of articulated 3D animal motions from raw, unlabeled online videos. Unlike existing approaches for 3D motion synthesis, our model requires no pose annotations or parametric shape models for training; it learns purely from a collection of unlabeled web video clips, leveraging semantic correspondences distilled from self-supervised image features. At the core of our method is a video Photo-Geometric Auto-Encoding framework that decomposes each training video clip into a set of explicit geometric and photometric representations, including a rest-pose 3D shape, an articulated pose sequence, and texture, with the objective of re-rendering the input video via a differentiable renderer. This decomposition allows us to learn a generative model over the underlying articulated pose sequences akin to a Variational Auto-Encoding (VAE) formulation, but without requiring any external pose annotations. At inference time, we can generate new motion sequences by sampling from the learned motion VAE, and create plausible 4D animations of an animal automatically within seconds given a single input image.

We introduce Temporal Residual Jacobians as a novel representation to enable data-driven motion transfer. Our approach does not assume access to any rigging or intermediate shape keyframes, produces geometrically and temporally consistent motions, and can be used to transfer long motion sequences. Central to our approach are two dedicated neural networks that individually predict the local geometric and temporal changes that are subsequently integrated, spatially and temporally, to produce the final animated meshes. The two networks are jointly trained, complement each other in producing spatial and temporal signals, and are supervised directly with 3D positional information. During inference, in the absence of keyframes, our method essentially solves a motion extrapolation problem. We test our setup on diverse meshes (synthetic and scanned shapes) to demonstrate its effectiveness in generating realistic and natural-looking animations on unseen body shapes.

Visual layout plays a critical role in graphic design fields such as advertising, posters, and web UI design. The recent trend towards content-aware layout generation through generative models has shown promise, yet it often overlooks the semantic intricacies of layout design by treating it as a simple numerical optimization. To bridge this gap, we introduce PosterLlama, a network designed for generating visually and textually coherent layouts by reformatting layout elements into HTML code and leveraging the rich design knowledge embedded within large language models. Furthermore, we enhance the robustness of our model with a unique depth-based poster augmentation strategy. This ensures our generated layouts remain semantically rich but also visually appealing, even with limited data. Our extensive evaluations across several benchmarks demonstrate that PosterLlama outperforms existing methods in producing authentic and content-aware layouts. It supports an unparalleled range of conditions, including but not limited to unconditional layout generation, element conditional layout generation, layout completion, among others, serving as a highly versatile user manipulation tool.

We propose GroundUp, the first sketch-based ideation tool for 3D city massing of urban areas. Architects start by making designs that create balance between constructed masses and open spaces, but existing software is aimed at late-stage precise building details. GroundUp aims to allow architects to easily iterate earlier, as they try out and share their ideas, moving back and forth between 2D sketching and 3D. Inspired by feedback from architects and existing workflows, our user sketches the initial footprints of multiple buildings in a top-down view. Our proposed interface then enables them to preview and sketch in a second, perspective view. All the while, they can visualize and navigate a 3D model, inferred to match the two sketches, leading them to change or refine the designs. The model driving our interface has two main components. First, we propose a tailored depth prediction network for perspective sketches, which exploits the top-down view cues. Second, we integrate a state-of-the-art latent diffusion model to enable the generation of plausible-looking buildings. This diffusion model uses a heightfield representation, which allows us to construct the city ''from the ground up''. It is conditioned on the top-down sketch and partial depth cues derived from the perspective …

In this work we propose to combine the advantages of learning-based and combinatorial formalisms for 3D shape matching. While learning-based methods lead to state-of-the-art matching performance, they do not ensure geometric consistency, so that obtained matchings are locally non-smooth. On the contrary, axiomatic, optimisation-based methods allow to take geometric consistency into account by explicitly constraining the space of valid matchings. However, existing axiomatic formalisms do not scale to practically relevant problem sizes, and require user input for the initialisation of non-convex optimisation problems. We work towards closing this gap by proposing a novel combinatorial solver that combines a unique set of favourable properties: our approach (i) is initialisation free, (ii) is massively parallelisable and powered by a quasi-Newton method, (iii) provides optimality gaps, and (iv) delivers improved matching quality with decreased runtime and globally optimal results for many instances.

In this paper, we introduce a novel method called FRI-Net for 2D floorplan reconstruction from 3D point cloud. Existing methods typically rely on corner regression or box regression, which lack consideration for the global shapes of rooms. To address these issues, we propose a novel approach using a room-wise implicit representation to characterize the shapes of rooms in floorplans. By incorporating geometric priors of room layouts in floorplans into our training strategy, the generated room polygons are more geometrically regular. We conducted experiments on two challenging datasets, Structured3D and SceneCAD. Our method demonstrates improved performance compared to state-of-the-art methods, validating the effectiveness of our proposed representation for floorplan reconstruction.

Point clouds offer an attractive source of information to complement images in neural scene representations, especially when few images are available. Neural rendering methods based on point clouds do exist, but they do not perform well when the point cloud quality is low--e.g., sparse or incomplete, which is often the case with real-world data. We overcome these problems with a simple representation that aggregates point clouds at multiple scale levels with sparse voxel grids at different resolutions. To deal with point cloud sparsity, we average across multiple scale levels---but only among those that are valid, \ie, that have enough neighboring points in proximity to the ray of a pixel. To help model areas without points, we add a global voxel at the coarsest scale, thus unifying ``classical'' and point-based NeRF formulations. We validate our method on the NeRF Synthetic, ScanNet, and KITTI-360 datasets, outperforming the state of the art, with a significant gap compared to other NeRF-based methods, especially on more challenging scenes.

The inherent richness of geometric information in point cloud underscores the necessity of leveraging group equivariance, as preserving the topological structure of the point cloud up to the feature space provides an intuitive inductive bias for solving problems in 3D space. Since manifesting the symmetry by means of model architecture has an advantage over the dependence on the augmentation, it has been a crucial research topic in the point cloud field. However, existing methods have limitations in the non-continuity of groups or the complex architecture causing computational inefficiency. In this paper, we propose CSEConv: a novel point convolution layer equivariant under continuous SO(3) actions. Its structure is founded on the framework of group theory, realizing the convolution module defined on a sphere. Implementing its filters to be explicit, continuous, and rigorously equivariant functions defined upon the double coset space is the distinctive factor which makes our method more scalable than previous approaches. From the classification experiments on synthetic and real-world point cloud datasets, our method achieves the best accuracy, to the best of our knowledge, amidst point-based models equivariant against continuous rotation group.

Point cloud registration is a critical component in many vision-based applications, such as perception for autonomous systems. The registration of point cloud observations on a rigid object, or scene, amounts to estimating the rigid transformation relating them. However, in practical scenarios, these observations are often characterized by partial overlap as a result of being acquired from different viewpoints, as well as by different sampling patterns. In this paper, we adopt the Universal Manifold Embedding (UME) framework for the estimation of rigid transformations and extend it, so that it can accommodate scenarios involving partial overlap and differently sampled point clouds. UME is a methodology designed for mapping observations of the same object, related by rigid transformations, into a single low-dimensional linear subspace. This process yields a transformation-invariant representation of the observations, with its matrix form representation being covariant with the transformation. We extend the UME framework by introducing a UME-compatible feature extraction method augmented with a unique UME contrastive loss and a sampling equalizer. These components are integrated into a comprehensive and robust registration pipeline, named UMERegRobust. We propose the RotKITTI registration benchmark, specifically tailored to evaluate registration methods for scenarios involving large rotations. UMERegRobust achieves better than state-of-the-art performance on …

We propose FrePolad: frequency-rectified point latent diffusion, a point cloud generation pipeline integrating a variational autoencoder (VAE) with a denoising diffusion probabilistic model (DDPM) modeling the latent distribution. FrePolad simultaneously achieves high quality, diversity, and flexibility in point cloud cardinality for generation tasks while maintaining high computational efficiency. The improvement in generation quality and diversity is achieved through (1) a novel frequency rectification module via spherical harmonics designed to retain high-frequency content while learning the point cloud distribution; and (2) a latent DDPM to learn the regularized yet complex latent distribution. In addition, FrePolad supports variable point cloud cardinality by formulating the sampling of points as conditional distributions over a latent shape distribution. Finally, the low-dimensional latent space encoded by the VAE contributes to FrePolad's fast and scalable sampling. Our quantitative and qualitative evaluations demonstrate the state-of-the-art performance of FrePolad in terms of quality, diversity, and computational efficiency.

Unsupervised domain adaptation (UDA) in 3D segmentation tasks presents a formidable challenge, primarily stemming from the sparse and unordered nature of point clouds. Especially for LiDAR point clouds, the domain discrepancy becomes obvious across varying capture scenes, fluctuating weather conditions, and the diverse array of LiDAR devices in use. Inspired by the remarkable generalization capabilities exhibited by the vision foundation model, SAM, in the realm of image segmentation, our approach leverages the wealth of general knowledge embedded within SAM to unify feature representations across diverse 3D domains and further solves the 3D domain adaptation problem. Specifically, we harness the corresponding images associated with point clouds to facilitate knowledge transfer and propose an innovative hybrid feature augmentation methodology, which enhances the alignment between the 3D feature space and SAM's feature space, operating at both the scene and instance levels. Our method is evaluated on many widely-recognized datasets and achieves state-of-the-art performance.

Underwater image restoration is a challenging task because of strong water effects that increase dramatically with distance. This is worsened by lack of ground truth data of clean scenes without water. Diffusion priors have emerged as strong image restoration priors. However, they are often trained with a dataset of the desired restored output, which is not available in our case. To overcome this critical issue, we show how to leverage in-air images to train diffusion priors for underwater restoration. We also observe that only color data is insufficient, and augment the prior with a depth channel. We train an unconditional diffusion model prior on the joint space of color and depth, using standard RGBD datasets of natural outdoor scenes in air. Using this prior together with a novel guidance method based on the underwater image formation model, we generate posterior samples of clean images, removing the water effects. Even though our prior did not see any underwater images during training, our method outperforms state-of-the-art baselines for image restoration on very challenging scenes.

We revisit the problem of camera relocalization under memory budget through a combination of product quantization and map compression. This achieves high compression rates but leads to performance drop. To address memory performance tradeoff, we train a light-weight scene-specific auto-encoder network that performs quantization-dequantization in an end-to-end differentiable manner updating both product quantization centroids and network parameters. Unlike standard L2 reconstruction loss for training auto-encoder network, we show that additional margin-based metric losses are key to achieve good performance. Results show that for a descriptor memory of 1 MB, we can achieve competitive performance on Aachen Day with only 5 % drop in performance.

Recent advances in Neural Fields mostly rely on developing task-specific supervision which often complicates the models. Rather than developing hard-to-combine and specific modules, another approach generally overlooked is to directly inject generic priors on the scene representation (also called inductive biases) into the NeRF architecture. Based on this idea, we propose the RING-NeRF architecture which includes two inductive biases : a continuous multi-scale representation of the scene and an invariance of the decoder's latent space over spatial and scale domains. We also design a single reconstruction process that takes advantage of those inductive biases and experimentally demonstrates on-par performances in terms of quality with dedicated architecture on multiple tasks (anti-aliasing, few view reconstruction, SDF reconstruction without scene-specific initialization) while being more efficient. Moreover, RING-NeRF has the distinctive ability to dynamically increase the resolution of the model, opening the way to adaptive reconstruction.

Although time-of-flight (ToF) cameras are becoming the sensor-of-choice for numerous 3D imaging applications in robotics, augmented reality (AR) and human-computer interfaces (HCI), they do not explicitly consider scene or camera motion. Consequently, current ToF cameras do not provide 3D motion information, and the estimated depth and intensity often suffers from significant motion artifacts in dynamic scenes. In this paper, we propose a novel ToF imaging method for dynamic scenes, with the goal of simultaneously estimating 3D geometry, intensity, and 3D motion using a single indirect ToF (I-ToF) camera. Our key observation is that we can estimate 3D motion, as well as motion artifact-free depth and intensity by designing optical-flow-like algorithms that operate on coded correlation images captured by an I-ToF camera. Through the integration of a multi-frequency I-ToF approach with burst imaging, we demonstrate high-quality all-in-one (3D geometry, intensity, 3D motion) imaging even in challenging low signal-to-noise ratio scenarios. We show the effectiveness of our approach through thorough simulations and real experiments conducted across a wide range of motion and imaging scenarios, including indoor and outdoor dynamic scenes.

Phase shifting profilometry (PSP) is favored in high-precision 3D scanning due to its high accuracy, robustness, and pixel-wise property. However, a fundamental assumption of PSP that the object should remain static is violated in dynamic measurement, making PSP susceptible to object moving, resulting in ripple-like errors in the point clouds. We propose a pixel-wise and frame-wise loopable binomial self-compensation (BSC) algorithm to effectively and flexibly eliminate motion error in the four-step PSP. Our mathematical model demonstrates that by summing successive motion-affected phase frames weighted by binomial coefficients, motion error exponentially diminishes as the binomial order increases, enabling automatic error compensation through the motion-affected phase sequence, without the assistance of any intermediate variable. We demonstrate that our BSC outperforms the existing methods in reducing motion error, while achieving a depth map frame rate equal to the camera's acquisition rate (90 fps), enabling high-accuracy 3D reconstruction with a quasi-single-shot frame rate.

Non-line-of-sight (NLoS) reconstruction, i.e., the task of imaging scenes beyond the camera's field of view, is often implemented using source-and-sensor systems that scan the visible region and analyze secondary reflections of light that has interacted with the hidden static scene. Estimating human activity around the corner will be a task of major interest for emerging NLoS applications, and some attempts have been reported in the recent literature. However, due to the long exposure times and comprehensive scans needed for NLoS sensing, the reconstruction of continuous movement remains prone to artifacts and is unreliable. In this paper, we analyze the interplay between dynamic scenes and scanning hardware to identify possible failure cases for filtering and data-driven approaches. Our studies indicate that existing reconstruction methods are prone to systematic error due to the space-time skew introduced by scanning setups. To alleviate this issue, we propose an image formation model for dynamic scenes that explicitly integrates motion skew. Using this model, we construct a baseline method for human pose estimation that achieves high accuracy, even at very slow scan rates.

Synchronization involves the task of inferring unknown vertex values (belonging to a group) in a graph, from edges labeled with vertex relations. While many matrix groups (e.g., rotations or permutations) have received extensive attention in Computer Vision, a complete solution for projectivities is lacking. Only the 3x3 case has been addressed so far, by mapping the problem onto the Special Linear Group, but the 4x4 projective case has remained unexplored and is the focus here. We propose novel strategies to address this task, and demonstrate their effectiveness in synthetic experiments, as well as on an application to projective Structure from Motion.

We address the semi-calibrated relative pose estimation problem where we assume the principal point to be in the center of the image and estimate the focal lengths, relative rotation, and translation of two cameras. We introduce the first minimal solver that requires only a single affine correspondence in conjunction with predicted monocular depth. Recognizing its degeneracy when the correspondence stems from a fronto-parallel plane, we present an alternative solver adept at recovering the correct solution under such circumstances. By integrating these methods within the GC-RANSAC framework, we show they surpass standard approaches, delivering more accurate poses at comparable runtimes across large-scale, publicly available indoor and outdoor datasets. The code will be public.

The extraction of keypoints in images is at the basis of many computer vision applications, from localization to 3D reconstruction. Keypoints come with a score permitting to rank them according to their quality. While learned keypoints often exhibit better properties than handcrafted ones, their scores are not easily interpretable, making it virtually impossible to compare the quality of individual keypoints across methods. We propose a framework that can refine, and at the same time characterize with an interpretable score, the keypoints extracted by any method. Our approach leverages a modified robust Gaussian Mixture Model fit designed to both reject non-robust keypoints and refine the remaining ones. Our score comprises two components: one relates to the probability of extracting the same keypoint in an image captured from another viewpoint, the other relates to the localization accuracy of the keypoint. These two interpretable components permit a comparison of individual keypoints extracted across different methods. Through extensive experiments we demonstrate that, when applied to popular keypoint detectors, our framework consistently improves the repeatability of keypoints as well as their performance in homography and two/multiple-view pose recovery tasks.

Simultaneous Localization and Mapping (SLAM) has been crucial across various domains, including autonomous driving, mobile robotics, and mixed reality. Dense visual SLAM, leveraging RGB-D camera systems, offers advantages but faces challenges in achieving real-time performance, robustness, and scalability for large-scale scenes. Recent approaches utilizing neural implicit scene representations show promise but suffer from high computational costs and memory requirements. ESLAM introduced a plane-based tensor decomposition but still struggled with memory growth. Addressing these challenges, we propose a more efficient visual SLAM model, called LRSLAM, utilizing low-rank tensor decomposition methods. Our approach, leveraging the Six-axis and CP decompositions, achieves better convergence rates, memory efficiency, and reconstruction/localization quality than existing state-of-the-art approaches. Evaluation across diverse indoor RGB-D datasets demonstrates LRSLAM's superior performance in terms of parameter efficiency, processing time, and accuracy, retaining reconstruction and localization quality. Our code will be publicly available upon publication.

Two-view pose estimation is essential for map-free visual relocalization and object pose tracking tasks. However, traditional matching methods suffer from time-consuming robust estimators, while deep learning-based pose regressors only cater to camera-to-world pose estimation, lacking generalizability to different image sizes and camera intrinsics. In this paper, we propose SRPose, a sparse keypoint-based framework for two-view relative pose estimation in camera-to-world and object-to-camera scenarios. SRPose consists of a sparse keypoint detector, an intrinsic-calibration position encoder, and promptable prior knowledge-guided attention layers. Given two RGB images of a fixed scene or a moving object, SRPose estimates the relative camera or 6D object pose transformation. Extensive experiments demonstrate that SRPose achieves competitive or superior performance compared to state-of-the-art methods in terms of accuracy and speed, showing generalizability to both scenarios. It is robust to varying different image sizes and camera intrinsics, and can be deployed with low computing resources.

Object pose distribution estimation is crucial in robotics for better path planning and handling of symmetric objects. Recent distribution estimation approaches employ contrastive learning-based approaches by maximizing the likelihood of a single pose estimate in the absence of a CAD model. We propose a pose distribution estimation method leveraging symmetry respecting correspondence distributions and shape information obtained using a CAD model. Contrastive learning-based approaches require an exhaustive amount of training images from different viewpoints to learn the distribution properly, which is not possible in realistic scenarios. Instead, we propose a pipeline that can leverage correspondence distributions and shape information from the CAD model, which are later used to learn pose distributions. Besides, having access to pose distribution based on correspondences before learning pose distributions conditioned on images, can help formulate the loss between distributions. The prior knowledge of distribution also helps the network to focus on getting sharper modes instead. With the CAD prior, our approach converges much faster and learns distribution better by focusing on learning sharper distribution near all the valid modes, unlike contrastive approaches, which focus on a single mode at a time. We achieve benchmark results on SYMSOL-I and T-Less datasets.

Rigid and articulated objects are common in our daily lives. Pose estimation tasks for both types of objects have been extensively studied within their respective domains. However, a universal framework capable of estimating the pose of both rigid and articulated objects has yet to be reported. In this paper, we introduce a Universal 9D Category-level Object Pose Estimation (U-COPE) framework, designed to address this gap. Our approach offers a novel perspective on rigid and articulated objects, redefining their pose estimation problems to unify them into a common task. Leveraging either 3D point cloud or RGB-D image inputs, we extract Point Pair Features ~(PPF) independently from each object part for end-to-end learning. Moreover, instead of direct prediction as seen in prior art, we employ a universal voting strategy to derive decisive parameters crucial for object pose estimation. Our network is trained end-to-end to optimize three key objectives: Joint Information, Part Segmentation, and 9D pose estimation through parameter voting. Extensive experiments validate the robustness of our method in estimating poses for both rigid and articulated objects, which demonstrates the generalizability to unseen object instances, too. Notably, our approach achieves state-of-the-art performance on synthetic datasets and real-world datasets. Our code will be publicly …

We present EgoPoseFormer, a simple yet effective transformer-based model for stereo egocentric human pose estimation. The main challenge in egocentric pose estimation is overcoming joint invisibility, which is caused by self-occlusion or a limited field of view (FOV) of head-mounted cameras. Our approach overcomes this challenge by incorporating a two-stage pose estimation paradigm: in the first stage, our model leverages the global information to estimate each joint’s coarse location, then in the second stage, it employs a DETR style transformer to refine the coarse locations by exploiting fine-grained stereo visual features. In addition, we present a deformable stereo attention operation to enable our transformer to effectively process multi-view features, which enables it to accurately localize each joint in the 3D world. We evaluate our method on the stereo UnrealEgo dataset and show it significantly outperforms previous approaches while being computationally efficient: it improves MPJPE by 27.4mm (45% improvement) with only 7.9% model parameters and 13.1% FLOPs compared to the state-of-the-art. Surprisingly, with proper training techniques, we find that even our first-stage pose proposal network can achieve superior performance compared to previous arts. We also show that our method can be seamlessly extended to monocular settings, which achieves state-of-the-art performance on …

We present Multi-HMR, a strong single-shot model for multi-person 3D human mesh recovery from a single RGB image. Predictions encompass the whole body, i.e., including hands and facial expressions, using the SMPL-X parametric model and 3D location in the camera coordinate system. Our model detects people by predicting coarse 2D heatmaps of person locations, using features produced by a standard Vision Transformer (ViT) backbone. It then predicts their whole-body pose, shape and 3D location using a new cross-attention module called the Human Prediction Head (HPH), with one query attending to the entire set of features for each detected person. As direct prediction of fine-grained hands and facial poses in a single shot, i.e., without relying on explicit crops around body parts, is hard to learn from existing data, we introduce CUFFS, the Close-Up Frames of Full-body Subjects dataset, containing humans close to the camera with diverse hand poses. We show that incorporating this dataset into training further enhances predictions, particularly for hands, enabling us to achieve competitive performance. Multi-HMR also optionally accounts for camera intrinsics, if available, by encoding camera ray directions for each image token. This simple design achieves strong performance on whole-body and body-only benchmarks simultaneously. We train …

Recent pose-based gait recognition methods, which utilize human skeletons as the model input, have demonstrated significant potential in handling variations in clothing and occlusions. However, methods relying on such skeleton to encode pose are constrained mainly by two problems: (1) poor performance caused by the shape loss, and (2) lack of generalizability. Addressing these limitations, we revisit pose-based gait recognition and develop GaitHeat, a heatmap-based framework that largely enhances performance and robustness by utilizing a new modality to encode pose rather than keypoint coordinates. We make our efforts from two aspects, the pipeline and the extraction of multi-channel heatmap features. Specifically, the process of resizing and centering is performed in the RGB space to largely preserve the integrity of heatmap information. To boost the generalization across various datasets further, we propose a pose-guided heatmap alignment module to eliminate the influence of gait-irrelevant covariates. Furthermore, a global-local network incorporating an efficient fusion branch is designed to improve the extraction of semantic information. Compared to skeleton-based methods, GaitHeat exhibits superior performance in learning gait features and demonstrates effective generalization across different datasets. Experiments on three datasets reveal that our proposed method achieves state-of-the-art results for pose-based gait recognition, comparable to that of …

In this study, we investigate the effectiveness of synthetic data in enhancing egocentric hand-object interaction detection. Via extensive experiments and comparative analyses on three egocentric datasets, VISOR, EgoHOS, and ENIGMA-51, our findings reveal how to exploit synthetic data for the HOI detection task when real labeled data are scarce or unavailable. Specifically, by leveraging only 10% of real labeled data, we achieve improvements in Overall AP compared to baselines trained exclusively on real data of: +5.67% on EPIC-KITCHENS VISOR, +8.24% on EgoHOS, and +11.69% on ENIGMA-51. Our analysis is supported by a novel data generation pipeline and the newly introduced HOI-Synth benchmark which augments existing datasets with synthetic images of hand-object interactions automatically labeled with hand-object contact states, bounding boxes, and pixel-wise segmentation masks. Data, code, and data generation tools to support future research are released at: https://fpv-iplab.github.io/HOI-Synth/.

Although hands frequently exhibit motion blur due to their dynamic nature, existing approaches for 3D hand recovery often disregard the impact of motion blur in hand images. Blurry hand images contain hands from multiple time steps, lacking precise hand location at a specific time step and introducing temporal ambiguity, leading to multiple possible hand trajectories. To address this issue and in the absence of datasets with real blur, we introduce the EBH dataset, which provides 1) hand images with real motion blur and 2) event data for authentic representation of fast hand movements. In conjunction with our new dataset, we present EBHNet, a novel network capable of recovering 3D hands from diverse input combinations, including blurry hand images, events, or both. Here, the event stream enhances motion understanding in blurry hands, addressing temporal ambiguity. Recognizing that blurry hand images include not only single 3D hands at a time step but also multiple hands along their motion trajectories, we design EBHNet to generate 3D hand sequences in motion. Moreover, to enable our EBHNet to predict 3D hands at novel, unsupervised time steps using a single shared module, we employ a Transformer-based module, temporal splitter, into EBHNet. Our experiments show the superior …

Existing datasets for 3D hand-object interaction are limited either in the data cardinality, data variations in interaction scenarios, or the quality of annotations. In this work, we present a comprehensive new training dataset for hand-object interaction called HOGraspNet. It is the only real dataset that captures full grasp taxonomies, providing grasp annotation and wide intraclass variations. Using grasp taxonomies as atomic actions, their space and time combinatorial can represent complex hand activities around objects. We select 22 rigid objects from the YCB dataset and 8 other compound objects using shape and size taxonomies, ensuring coverage of all hand grasp configurations. The dataset includes diverse hand shapes from 99 participants aged 10 to 74, continuous video frames, and a 1.5M RGB-Depth of sparse frames with annotations. It offers labels for 3D hand and object meshes, 3D keypoints, contact maps, and \emph{grasp labels}. Accurate hand and object 3D meshes are obtained by fitting the hand parametric model (MANO) and the hand implicit function (HALO) to multi-view RGBD frames, with the MoCap system only for objects. Note that HALO fitting does not require any parameter tuning, enabling scalability to the dataset's size with comparable accuracy to MANO. We evaluate HOGraspNet on relevant tasks: …

Reaching-and-grasping is a fundamental skill for robotic ma-nipulation, but existing methods usually train models on a specific gripper and cannot be reused on another gripper without retraining. In this paper, we propose a novel method that can learn a unifed policy modelthat can be easily transferred to different dexterous grippers. Our method consists of two stages: a gripper-agnostic policy model that predicts thedisplacements of predefined key points on the gripper, and a gripperspecifc adaptation model that translates these displacements into adjustments for controlling the grippers’ joints.The gripper state and interactions with objects are captured at the finger level using robust geometric representations integrated with a transformer-based network to address variations in gripper morphology and geometry. We evaluateour method on several dexterous grippers and objects of diverse shapes.And the result shows our method signifcantly outperforms the base-line methods. Our method pioneers the transfer of grasp policies acrossdifferent dexterous grippers, and demonstrates the potential of learn-ing generalizable and transferable manipulation skills for various robotichands.

We propose a novel drone application under real-world scenarios – free-viewpoint rendering of outdoor sports scenes, including the dynamic athlete and the 360° background. Outdoor sports have long-range human motions and large-scale scene structures which make the task rather challenging. Existing methods either rely on dense camera arrays which costs much, or a handheld moving camera which struggles to handle real sports scenes. We build a novel drone-based system using an RGB camera to reconstruct the 4D dynamic human along with the 3D unbounded scene, rendering free-viewpoint videos at any time. We also propose submodules for calibration and human motion capture, as a system-level design for improved robustness and efficiency. We collect a dataset AerialRecon and conduct extensive experiments on real-world scenarios. Compared with existing SOTA systems, our system demonstrates superior performance and applicability to real-world outdoor sports scenes.

Current exposure correction methods encounter three challenges, labor-intensive paired data annotation, limited generalizability, and performance degradation in low-level computer vision tasks. In this work, we introduce an innovative Unsupervised Exposure Correction (UEC) method that eliminates the need for manual annotations, offers improved generalizability, and enhances performance in low-level downstream tasks. Our model is trained using freely available paired data from an emulated Image Signal Processing (ISP) pipeline. This approach does not need expensive manual annotations, thereby minimizing individual style biases from the annotation and consequently improving its generalizability. Furthermore, we present a large-scale Radiometry Correction Dataset, specifically designed to emphasize exposure variations, to facilitate unsupervised learning. In addition, we develop a transformation function that preserves image details and outperforms state-of-the-art supervised methods, while utilizing only 0.01% of their parameter count. Our work further investigates the broader impact of exposure correction on downstream tasks, including edge detection, demonstrating its effectiveness in mitigating the adverse effects of poor exposure on low-level features. The source code and dataset will be made publicly available.

Learning a self-supervised Monocular Depth Estimation (MDE) model with great generalization remains significantly challenging. Despite the success of adversarial augmentation in the supervised learning generalization, naively incorporating it into self-supervised MDE methods potentially causes over-regularization, suffering from severe performance degradation. In this paper, we conduct qualitative analysis and illuminate the main causes: (i) inherent sensitivity in the UNet-alike depth network and (ii) dual optimization conflict caused by over-regularization. To tackle these issues, we propose a general adversarial training framework, named Stabilized Conflict-optimization Adversarial Training (SCAT), integrating adversarial data augmentation into self-supervised MDE methods to achieve a balance between stability and generalization. Specifically, we devise an effective Scaling Depth Network that tunes the coefficients of long skip connection in DepthNet and effectively stabilizes the training process. Then, we propose a Conflict Gradient Surgery strategy, which progressively integrates the adversarial gradient and optimizes the model toward a conflict-free direction. Extensive experiments on five benchmarks demonstrate that SCAT can achieve state-of-the-art performance and significantly improve the generalization capability of existing self-supervised MDE methods.

In this paper, we present a learning-based framework for improved sparse depth video completion. Given a sparse depth map and a color image, our approach makes a cost volume of a certain viewpoint that constructed on depth hypothesis planes. To effectively handle sequential cost volumes of the multiple viewpoints, we introduce a learning-based cost volume fusion framework, namely RayFusion, that effectively leverages the attention mechanism for each pair of overlapped rays in cost volumes. As a result of leveraging feature statistics accumulated over time, our proposed framework consistently outperforms or rivals state-of-the-art approaches on diverse indoor and outdoor datasets, including the KITTI Depth Completion benchmark, VOID Depth Completion benchmark, and ScanNetV2 dataset, using 94.5% fewer network parameters than the state-of-the-art approach, LRRU.

This paper introduces PatchRefiner, an advanced framework for metric single image depth estimation aimed at high-resolution real-domain inputs. While depth estimation is crucial for applications such as autonomous driving, 3D generative modeling, and 3D reconstruction, achieving accurate high-resolution depth in real-world scenarios is challenging due to the constraints of existing architectures and the scarcity of detailed real-world depth data. PatchRefiner adopts a tile-based methodology, reconceptualizing high-resolution depth estimation as a refinement process, which results in notable performance enhancements. Utilizing a pseudo-labeling strategy that leverages synthetic data, PatchRefiner incorporates a Detail and Scale Disentangling (DSD) loss to enhance detail capture while maintaining scale accuracy, thus facilitating the effective transfer of knowledge from synthetic to real-world data. Our extensive evaluations demonstrate PatchRefiner's superior performance, significantly outperforming existing benchmarks on the Unreal4KStereo dataset by 18.1% in terms of the root mean squared error (RMSE) and showing marked improvements in detail accuracy and consistent scale estimation on diverse real-world datasets like CityScape, ScanNet++, and ETH3D.

High frame rate and accurate depth estimation plays an important role in several tasks crucial to robotics and automotive perception. To date, this can be achieved through ToF and LiDAR devices for indoor and outdoor applications, respectively. However, their applicability is limited by low frame rate, energy consumption, and spatial sparsity. Depth on Demand (DoD) allows for accurate temporal and spatial depth densification achieved by exploiting a high frame rate RGB sensor coupled with a potentially lower frame rate and sparse active depth sensor. Our proposal jointly enables lower energy consumption and denser shape reconstruction by significantly reducing the streaming requirements on the depth sensor. We present extended evidence assessing the effectiveness of DoD on indoor and outdoor video datasets, covering both environment scanning and automotive perception use cases.

Self-driving vehicles (SDVs) require accurate calibration of LiDARs and cameras to fuse sensor data accurately for autonomy. Traditional calibration methods typically leverage fiducials captured in a controlled and structured scene and compute correspondences to optimize over. These approaches are costly and require substantial infrastructure and operations, making it challenging to scale for vehicle fleets. In this work, we propose UniCal, a unified framework for effortlessly calibrating SDVs equipped with multiple LiDARs and cameras. Our approach is built upon a differentiable scene representation capable of rendering multi-view geometrically and photometrically consistent sensor observations. We jointly learn the sensor calibration and the underlying scene representation through differentiable volume rendering, utilizing outdoor sensor data without the need for specific calibration fiducials. This "drive-and-calibrate" approach significantly reduces costs and operational overhead compared to existing calibration systems, enabling efficient calibration for large SDV fleets at scale. To ensure geometric consistency across observations from different sensors, we introduce a novel surface alignment loss that combines feature-based registration with neural rendering, as well as a coarse-to-fine sampling approach to optimize regions of interest for sensor alignment. Comprehensive evaluations on multiple datasets demonstrate that UniCal outperforms or matches the accuracy of existing calibration approaches while being more efficient, …

Multi-modal crowd counting involves estimating crowd density from both visual and thermal/depth images. This task is challenging due to the significant gap between these distinct modalities. In this paper, we propose a novel approach by introducing an auxiliary broker modality and on this basis frame the task as a triple-modal learning problem. We devise a fusion-based method to generate this broker modality, leveraging a non-diffusion, lightweight counterpart of modern denoising diffusion-based fusion models. Additionally, we identify and address the ghosting effect caused by direct cross-modal image fusion in multi-modal crowd counting. Through extensive experimental evaluations on popular multi-modal crowd counting datasets, we demonstrate the effectiveness of our method, which introduces only 4 million additional parameters, yet achieves promising results. We will release the source code upon the acceptance of the paper.

Accurate depth information is crucial for enhancing the performance of multi-view 3D object detection. Despite the success of some existing multi-view 3D detectors utilizing pixel-wise depth supervision, they overlook two significant phenomena: 1) the depth supervision obtained from LiDAR points is usually distributed on the surface of the object, which is not so friendly to existing DETR-based 3D detectors due to the lack of the depth of 3D object center; 2) for distant objects, fine-grained depth estimation of the whole object is more challenging. Therefore, we argue that the object-wise depth (or 3D center of the object) is essential for accurate detection. In this paper, we propose a new multi-view 3D object detector named OPEN, whose main idea is to effectively inject object-wise depth information into the network through our proposed object-wise position embedding. Specifically, we first employ an object-wise depth encoder, which takes the pixel-wise depth map as a prior, to accurately estimate the object-wise depth. Then, we utilize the proposed object-wise position embedding to encode the object-wise depth information into the transformer decoder, thereby producing 3D object-aware features for final detection. Extensive experiments verify the effectiveness of our proposed method. Furthermore, OPEN achieves a new state-of-the-art performance with …

Although multi-view 3D object detection based on the Bird's-Eye-View (BEV) paradigm has garnered widespread attention as an economical and deployment-friendly perception solution for autonomous driving, there is still a performance gap compared to LiDAR-based methods. In recent years, several cross-modal distillation methods have been proposed to transfer beneficial information from teacher models to student models, with the aim of enhancing performance. However, these methods face challenges due to discrepancies in feature distribution originating from different data modalities and network structures, making knowledge transfer exceptionally challenging. In this paper, we propose a Foreground Self-Distillation (FSD) scheme that effectively avoids the issue of distribution discrepancies, maintaining remarkable distillation effects without the need for pre-trained teacher models or cumbersome distillation strategies. Additionally, we design two Point Cloud Intensification (PCI) strategies to compensate for the sparsity of point clouds by frame combination and pseudo point assignment. Finally, we develop a Multi-Scale Foreground Enhancement (MSFE) module to extract and fuse multi-scale foreground features by predicted elliptical Gaussian heatmap, further improving the model's performance. We integrate all the above innovations into a unified framework named FSD-BEV. Extensive experiments on the nuScenes dataset exhibit that FSD-BEV achieves state-of-the-art performance, highlighting its effectiveness.

The visual detection and tracking of surface terrain is required for spacecraft to safely land on or navigate within close proximity to celestial objects. Current approaches rely on template matching with pre-gathered patch-based features, which are expensive to obtain and a limiting factor in perceptual capability. While recent literature has focused on in-situ detection methods to enhance navigation and operational autonomy, robust description is still needed. In this work, we explore metric learning as the lightweight feature description mechanism and find that current solutions fail to address inter class similarity and multi-view observational geometry. We attribute this to the view-unaware attention mechanism and introduce Multi-view Attention Regularizations (MARs) to constrain the channel and spatial attention across multiple feature views, regularizing the what and where of attention focus. We thoroughly analyze many modern metric learning losses with and without MARs and demonstrate improved terrain-feature recognition performance by upwards of 85%. We additionally introduce the Luna-1 dataset, consisting of Moon crater landmarks and reference navigation frames from NASA mission data to support future research in this difficult task. Luna-1 and source code are publicly available at https://droneslab.github.io/mars/.

Radar-based perception has gained increasing attention in autonomous driving, yet the inherent sparsity of radars poses challenges. Radar raw data often contains excessive noise, whereas radar point clouds retain only limited information. In this work, we holistically treat the sparse nature of radar data by introducing an adaptive subsampling method together with a tailored network architecture that exploits the sparsity patterns to discover global and local dependencies in the radar signal. Our subsampling module selects a subset of pixels from range-doppler (RD) spectra that contribute most to the downstream perception tasks. To improve the feature extraction on sparse subsampled data, we propose a new way of applying graph neural networks on radar data and design a novel two-branch backbone to capture both global and local neighbor information. An attentive fusion module is applied to combine features from both branches. Experiments on the RADIal dataset show that our SparseRadNet exceeds state-of-the-art (SOTA) performance in object detection and achieves close to SOTA accuracy in freespace segmentation, meanwhile using sparse subsampled input data.

Masked Autoencoders (MAE) play a pivotal role in learning potent representations, delivering outstanding results across various 3D perception tasks essential for autonomous driving. In real-world driving scenarios, it's commonplace to deploy multiple sensors for comprehensive environment perception. Despite integrating multi-modal features from these sensors can produce rich and powerful features, there is a noticeable challenge in MAE methods addressing this integration. This research delves into multi-modal Masked Autoencoders tailored for a unified representation space in autonomous driving, aiming to pioneer a more efficient fusion of two distinct modalities. To intricately marry the semantics inherent in images with the geometric intricacies of LiDAR point clouds, we propose UniM^2AE. This model stands as a potent yet straightforward, multi-modal self-supervised pre-training framework, mainly consisting of two designs. First, it projects the features from both modalities into a cohesive 3D volume space, ingeniously expanded from the bird's eye view (BEV) to include the height dimension. The extension allows for a precise representation of objects and reduces information loss when aligning multi-modal features. Second, the Multi-modal 3D Interactive Module (MMIM) is invoked to facilitate the efficient inter-modal interaction during the interaction process. Extensive experiments conducted on the nuScenes Dataset attest to the efficacy of UniM^2AE, …

The tasks of object detection and trajectory forecasting play a crucial role in understanding the scene for autonomous driving. These tasks are typically executed in a cascading manner, making them prone to compounding errors. Furthermore, there is usually a very thin interface between the two tasks, creating a lossy information bottleneck. To address these challenges, our approach formulates the union of the two tasks as a trajectory refinement problem, where the first pose is the detection (current time), and the subsequent poses are the waypoints of the multiple forecasts (future time). To tackle this unified task, we design a refinement transformer that infers the presence, pose, and multi-modal future behaviors of objects directly from LiDAR point clouds and high-definition maps. We call this model DeTra, short for object DEtection and TRAjectory forecasting. In our experiments, we observe that DeTra outperforms the state-of-the-art on Argoverse 2 Sensor and Waymo Open Dataset by a large margin, across a broad range of metrics. Last but not least, we perform extensive ablation studies that show the value of refinement for this task, that every proposed component contributes positively to its performance, and that key design choices were made.

This paper aims to tackle the problem of modeling dynamic urban streets for autonomous driving scenes. Recent methods extend NeRF by incorporating tracked vehicle poses to animate vehicles, enabling photo-realistic view synthesis of dynamic urban street scenes. However, significant limitations are their slow training and rendering speed. We introduce Street Gaussians, a new explicit scene representation that tackles these limitations. Specifically, the dynamic urban scene is represented as a set of point clouds equipped with semantic logits and 3D Gaussians, each associated with either a foreground vehicle or the background. To model the dynamics of foreground object vehicles, each object point cloud is optimized with optimizable tracked poses, along with a 4D spherical harmonics model for the dynamic appearance. The explicit representation allows easy composition of object vehicles and background, which in turn allows for scene editing operations and rendering at 135 FPS (1066 * 1600 resolution) within half an hour of training. The proposed method is evaluated on multiple challenging benchmarks. Experiments show that Street Gaussians consistently outperforms state-of-the-art methods across all datasets. The code will be released to ensure reproducibility.

In this paper, we introduce PredBench, a benchmark tailored for the holistic evaluation of spatio-temporal prediction networks. Despite significant progress in this field, there remains a lack of a standardized framework for a detailed and comparative analysis of various prediction network architectures. PredBench addresses this gap by conducting large-scale experiments, upholding standardized and appropriate experimental settings, and implementing multi-dimensional evaluations. This benchmark integrates 12 widely adopted methods with 15 diverse datasets across multiple application domains, offering extensive evaluation of contemporary spatio-temporal prediction networks. Through meticulous calibration of prediction settings across various applications, PredBench ensures evaluations relevant to their intended use and enables fair comparisons. Moreover, its multi-dimensional evaluation framework broadens the analysis with a comprehensive set of metrics, providing deep insights into the capabilities of models. The findings from our research offer strategic directions for future developments in the field. Further, we will release our extensive codebase to encourage and support continued advancements in spatio-temporal prediction research.

Semantic segmentation empowers numerous real-world applications, such as autonomous driving and augmented/mixed reality. These applications often operate on high-resolution images (e.g., 8 megapixels) to capture the fine details. However, this comes at the cost of considerable computational complexity, hindering the deployment in latency-sensitive scenarios. In this paper, we introduce SparseRefine, a novel approach that enhances dense low-resolution predictions with sparse high-resolution refinements. Based on coarse low-resolution outputs, SparseRefine first uses an entropy selector to identify a sparse set of pixels with high entropy. It then employs a sparse feature extractor to efficiently generate the refinements for those pixels of interest. Finally, it leverages a gated ensembler to apply these sparse refinements to the initial coarse predictions. SparseRefine can be seamlessly integrated into any existing semantic segmentation model, regardless of CNN- or ViT-based. SparseRefine achieves significant speedup: 1.5 to 3.9 times when applied to HRNet-W48, SegFormer-B5, Mask2Former-T/L and SegNeXt-L on Cityscapes, with negligible to no loss of accuracy. We will release the code to reproduce our results. Our "dense+sparse'' paradigm paves the way for efficient high-resolution visual computing.

Vectorized high-definition (HD) maps contain detailed information about surrounding road elements, which are crucial for various downstream tasks in modern autonomous vehicles, such as motion planning and vehicle control. Recent works attempt to directly detect the vectorized HD map as a point set prediction task, achieving notable detection performance improvements. However, these methods usually overlook and fail to analyze the important inner-instance correlations between predicted points, impeding further advancements. To address this issue, we investigate the utilization of inner-instance information for vectorized high-definition mapping through transformers, and propose a powerful system named \textbf{InsMapper}, which effectively harnesses inner-instance information with three exquisite designs, including hybrid query generation, inner-instance query fusion, and inner-instance feature aggregation. The first two modules can better initialize queries for line detection, while the last one refines predicted line instances. InsMapper is highly adaptable and can be seamlessly modified to align with the most recent HD map detection frameworks. Extensive experimental evaluations are conducted on the challenging NuScenes and Argoverse 2 datasets, where InsMapper surpasses the previous state-of-the-art method, demonstrating its effectiveness and generality.

Autonomous vehicles rely extensively on perception systems to navigate and interpret their surroundings. Despite significant advancements in these systems recently, challenges persist under conditions like occlusion, extreme lighting, or in unfamiliar urban areas. Unlike these systems, humans do not solely depend on immediate observations to perceive the environment. In navigating new cities, humans gradually develop a preliminary mental map to supplement real-time perception during subsequent visits. Inspired by this human approach, we introduce a novel framework, Pre-Sight, that leverages past traversals to construct static prior memories, enhancing online perception in later navigations. Our method involves optimizing a city-scale neural radiance field with data from previous journeys to generate neural priors. These priors, rich in semantic and geometric details, are derived without manual annotations and can seamlessly augment various state-of-the-art perception models, improving their efficacy with minimal additional computational cost. Experimental results on the nuScenes dataset demonstrate the framework's high compatibility with diverse online perception models. Specifically, it shows remarkable improvements in HD-map construction and occupancy prediction tasks, highlighting its potential as a new perception framework for autonomous driving systems.

Embodied vision-based real-world systems, such as mobile robots, require careful balancing between energy consumption, compute latency, and safety constraints to optimize operation across dynamic tasks and contexts. As local computation tends to be restricted, offloading the computation, i.e., to a remote server, can save local resources while providing access to high-quality predictions from powerful and large models. Yet, the resulting communication and latency overhead has led to limited usability of cloud models in dynamic, safety-critical, real-time settings. Towards effectively addressing this trade-off, in this work, we introduce UniLCD, a novel hybrid inference framework for enabling flexible local-cloud collaboration. By efficiently optimizing a flexible routing module via reinforcement learning and a suitable multi-task objective, UniLCD is specifically designed to support multiple constraints of safety-critical end-to-end mobile systems. We validate the proposed approach using a challenging crowded navigation task requiring frequent and timely switching between local and cloud operations. UniLCD demonstrates both improved overall performance and efficiency, by over 17% compared to state-of-the-art baselines based on various split computing strategies.

Directly producing planning results from raw sensors has been a long-desired solution for autonomous driving and has attracted increasing attention recently. Most existing end-to-end autonomous driving methods factorize this problem into perception, motion prediction, and planning. However, we argue that the conventional progressive pipeline still cannot comprehensively model the entire traffic evolution process, e.g., the future interaction between the ego car and other traffic participants and the structural trajectory prior. In this paper, we explore a new paradigm for end-to-end autonomous driving, where the key is to predict how the ego car and the surroundings evolve given past scenes. We propose GenAD, a generative framework that casts autonomous driving into a generative modeling problem. We propose an instance-centric scene tokenizer that first transforms the surrounding scenes into map-aware instance tokens. We then employ a variational autoencoder to learn the future trajectory distribution in a structural latent space for trajectory prior modeling. We further adopt a temporal model to capture the agent and ego movements in the latent space to generate more effective future trajectories. GenAD finally simultaneously performs motion prediction and planning by sampling distributions in the learned structural latent space conditioned on the instance tokens and using the learned …

Multi-agent trajectory prediction is crucial to autonomous driving and understanding the surrounding environment. In recent years, learning-based approaches for multi-agent trajectory prediction, such as primarily relying on graph neural networks, graph transformers, and hypergraph neural networks, have demonstrated outstanding performance on real-world datasets. However, hypergraph transformer-based method for trajectory prediction is yet to be explored. Therefore, we present a MultiscAle Relational Transformer (MART) network for multi-agent trajectory prediction. MART is a hypergraph transformer architecture to consider individual and group behaviors in transformer machinery. The core module of MART is the encoder, comprising a Pair-wise Relational Transformer (PRT) and a Hyper Relational Transformer (HRT). The encoder extends the capabilities of a relational transformer by introducing HRT, which integrates hyperedge features into the transformer mechanism, promoting the attention weights to focus on group-wise relations. In addition, we propose an Adaptive Group Estimator (AGE) designed to infer complex group relations in real-world environments. Extensive experiments on three real-world datasets (NBA, SDD, and ETH-UCY) demonstrate that our method achieves state-of-the-art performance, enhancing ADE/FDE by 3.9%/11.8% on the NBA dataset.

A major challenge in autonomous vehicle research is agents behavior modeling, which has critical applications including constructing realistic and reliable simulations for off-board evaluation and motion forecasting of traffic agents for onboard planning. Motion prediction models trained via supervised learning have recently proven effective at modeling agents across many domains. However, these models are subject to distribution shift when deployed at test-time. In this work, we improve the reliability of agent behaviors by closed-loop fine-tuning of behavior models with reinforcement learning. We demonstrate that we can improve overall performance, as well as targeted metrics such as collision rate, on the Waymo Open Sim Agents challenge benchmark. We also introduce a novel policy evaluation ranking benchmark for directly evaluating the ability of sim agents to measure planning quality and demonstrate the effectiveness of our approach on this new benchmark.

Achieving 3D understanding of non-Lambertian objects is an important task with many useful applications, but most existing algorithms struggle to deal with such objects. One major obstacle towards progress in this field is the lack of holistic non-Lambertian benchmarks---most benchmarks have low scene and object diversity, and none provide multi-layer 3D annotations for objects occluded by transparent surfaces. In this paper, we introduce LayeredFlow, a real world benchmark containing multi-layer ground truth annotation for optical flow of non-Lambertian objects. Compared to previous benchmarks, our benchmark exhibits greater scene and object diversity, with 15k high quality optical flow and stereo pairs taken over 185 indoor and outdoor scenes and 360 unique objects. Using LayeredFlow as evaluation data, we propose a new task called multi-layer optical flow. To provide training data for this task, we introduce a large-scale densely-annotated synthetic dataset containing 60k images within 30 scenes tailored for non-Lambertian objects. Training on our synthetic dataset enable model to predict multi-layer optical flow, while fine-tuning existing optical flow methods on the dataset notably boosts their performance on non-Lambertian objects without compromising the performance on diffuse objects.

Recent studies on motion estimation have advocated an optimized motion representation that is globally consistent across the entire video, preferably for every pixel. This is challenging as a uniform representation may not account for the complex and diverse motion and appearance of natural videos. We address this problem and propose a new test-time optimization method, named DecoMotion, for estimating per-pixel and long-range motion. DecoMotion explicitly decomposes video content into static scenes and dynamic objects, either of which uses a quasi-3D canonical volume to represent. DecoMotion separately coordinates the transformations between local and canonical spaces, facilitating an affine transformation for the static scene that corresponds to camera motion. For the dynamic volume, DecoMotion leverages discriminative and temporally consistent features to rectify the non-rigid transformation. The two volumes are finally fused to fully represent motion and appearance. This divide-and-conquer strategy leads to more robust tracking through occlusions and deformations and meanwhile obtains decomposed appearances. We conduct evaluations on the TAP-Vid benchmark. The results demonstrate our method boosts the point-tracking accuracy by a large margin and performs on par with some state-of-the-art dedicated point-tracking solutions. Code and models will be available upon acceptance of the paper.

Dynamic stereo matching is the task of estimating consistent disparities from stereo videos with dynamic objects. Recent learning-based methods prioritize optimal performance on a single stereo pair, resulting in temporal inconsistencies. Existing video methods apply per-frame matching and window-based cost aggregation across the time dimension, leading to low-frequency oscillations at the scale of the window size. Towards this challenge, we develop a bidirectional alignment mechanism for adjacent frames as a fundamental operation. We further propose a novel framework, BiDAStereo, that achieves consistent dynamic stereo matching. Unlike the existing methods, we model this task as local matching and global aggregation. Locally, we consider correlation in a triple-frame manner to pool information from adjacent frames and improve the temporal consistency. Globally, to exploit the entire sequence's consistency and extract dynamic scene cues for aggregation, we develop a motion-propagation recurrent unit. Extensive experiments demonstrate the performance of our method, showcasing improvements in prediction quality and achieving state-of-the-art results on various commonly used benchmarks. The code is available in supplemental material.

Leveraging the high temporal resolution of an event-based camera requires highly efficient event-by-event processing. However, dense prediction tasks require explicit pixel-level association, which is challenging for event-based processing frameworks. Existing works aggregate the events into a static frame-like representation at the cost of a much slower processing rate and high compute cost. To address this challenge, this work introduces an event-based spatiotemporal representation learning framework for efficiently solving dense prediction tasks. We uniquely handle the sparse, asynchronous events using an unstructured, set-based approach and project them into a hierarchically organized multi-level latent memory space that preserves the pixel-level structure. Low-level event streams are dynamically encoded into these latent structures through an explicit attention-based spatial association. Unlike existing works that update these memory stacks at a fixed rate, we introduce a data-adaptive update rate that recurrently keeps track of the past memory states and learns to update the corresponding memory stacks only when it has substantial new information, thereby improving the overall compute latency. Our method consistently achieves competitive performance across different event-based dense prediction tasks while ensuring much lower latency compared to the existing methods.

In low-light conditions, capturing videos with frame-based cameras often requires long exposure times, resulting in motion blur and reduced visibility. While frame-based motion deblurring and low-light enhancement have been studied, they still pose significant challenges. Event cameras have emerged as a promising solution for improving image quality in low-light environments and addressing motion blur. They provide two key advantages: capturing scene details well even in low light due to their high dynamic range, and effectively capturing motion information during long exposures due to their high temporal resolution. Despite efforts to tackle low-light enhancement and motion deblurring using event cameras separately, previous work has not addressed both simultaneously. To explore the joint task, we first establish real-world datasets for event-guided low-light enhancement and deblurring using a hybrid camera system based on beam splitters. Subsequently, we introduce an end-to-end framework to effectively handle these tasks. Our framework incorporates a module to efficiently leverage temporal information from events and frames. Furthermore, we propose a module to utilize cross-modal feature information to employ a low-pass filter for noise suppression while enhancing the main structural information. Our proposed method significantly outperforms existing approaches in addressing the joint task. Our project pages are available at https://github.com/intelpro/ELEDNet.

The ability to understand physical dynamics is critical for agents to act in the world. Here, we use Counterfactual World Modeling (CWM) to extract vision structures for dynamics understanding. CWM uses a temporally-factored masking policy for masked prediction of video data without annotations. This policy enables highly effective ``counterfactual prompting'' of the predictor, allowing a spectrum of visual structures to be extracted from a single pre-trained predictor in a zero-shot manner. We demonstrate that these structures are useful for physical dynamics understanding, allowing CWM to achieve the state-of-the-art performance on the Physion benchmark.

Hand motion prediction from both first- and third-person perspectives is vital for enhancing user experience in AR/VR and ensuring safe remote robotic arm control. Previous works typically focus on predicting hand motion trajectories or human body motion, with direct hand motion prediction remaining largely unexplored - despite the additional challenges posed by compact skeleton size. To address this, we propose a prompt-based Future Driven Diffusion Model (PromptFDDM) for predicting hand motion with guidance and prompts. Specifically, we develop a Spatial-Temporal Extractor Network (STEN) to predict hand motion with guidance, a Ground Truth Extractor Network (GTEN), and a Reference Data Generator Network (RDGN), which extract ground truth and substitute future data with generated reference data, respectively, to guide STEN. Additionally, interactive prompts generated from observed motions further enhance model performance. Experimental results on the FPHA and HO3D datasets demonstrate that the proposed PromptFDDM achieves state-of-the-art performance in both first- and third-person perspectives.

We introduce Nymeria - a large-scale, diverse, richly annotated human motion dataset collected in the wild with multi-modal egocentric devices. The dataset comes with a) full-body motion ground truth; b) egocentric multimodal recordings from Project Aria devices, including color, grayscale and eye-tracking cameras, IMUs, magnetometer, barometer, and multi-channel microphones; and c) an additional "observer" device providing a third-person viewpoint. We compute world-aligned 6DoF transformations for all sensors, across devices and capture sessions. The dataset also provides 3D scene point clouds and calibrated eye gaze. We derive a protocol to annotate hierarchical language descriptions of in-context human motion, from fine-grain dense body pose narrations, to simplified atomic actions and coarse activity summarization. To the best of our knowledge, Nymeria dataset is the world's largest collection of human motion in the wild with natural and diverse activities; first of its kind to provide synchronized and localized multi-device multimodal egocentric data; and also the world’s largest dataset of motion with language descriptions. It contains 1200 recordings of 300 hours daily activities from 264 participants 50 locations. The accumulated trajectory from participants is 399.2Km for the head and 1053.3Km for both wrists. To facilitate research, we define multiple research tasks in egocentric body tracking, …

Human motion generation stands as a significant pursuit in generative computer vision, while achieving long-sequence and efficient motion generation remains challenging. Recent advancements in state space models (SSMs), notably Mamba, have showcased considerable promise in long sequence modeling with an efficient hardware-aware design, which appears to be a promising direction to build motion generation model upon it. Nevertheless, adapting the SSMs to motion generation faces hurdles since the lack of specialized design architecture for modeling motion sequence. To address these multifaceted challenges, we introduce three key contributions. Firstly, we proposed Motion Mamba, an innovative yet straightforward approach that presents the pioneering motion generation model utilized SSMs. Secondly, we designed a Hierarchical Temporal Mamba (HTM) block to process temporal data by traversing through a symmetric architecture aimed at preserving motion consistency between frames. We also designed a Bidirectional Spatial Mamba (BSM) block to bidirectionally process latent poses, in order to enhance accurate motion generation within a temporal frame. Lastly, the proposed method has outperformed other well-established methods on the HumanML3D and KIT-ML datasets, which demonstrates strong capabilities of high-quality long sequence motion modeling and real-time human motion generation.

Controllable human motion synthesis is essential for applications in AR/VR, gaming and embodied AI. Existing methods often focus solely on either language or full trajectory control, lacking precision in synthesizing motions aligned with user-specified trajectories, especially for multi-joint control. To address these issues, we present TLControl, a novel method for realistic human motion synthesis, incorporating both low-level Trajectory and high-level Language semantics controls, through an integration of neural-based and optimization-based techniques. Specifically, we begin with training a VQ-VAE for a compact and well-structured latent motion space organized by body parts. We then propose a Masked Trajectories Transformer~(MTT) for predicting a motion distribution conditioned on language and trajectory. Once trained, we use MTT to sample initial motion predictions given user-specified partial trajectories and text descriptions as conditioning. Finally, we introduce a test-time optimization to refine these coarse predictions for precise trajectory control, which offers flexibility by allowing users to specify various optimization goals, and ensures high runtime efficiency. Comprehensive experiments show that TLControl significantly outperforms the state-of-the-art in trajectory accuracy and time efficiency, making it practical for interactive and high-quality animation generation.

We study a challenging task: text-to-motion synthesis, aiming to generate motions that align with textual descriptions and exhibit coordinated movements. Currently, the part-based methods introduce part partition into the motion synthesis process to achieve finer-grained generation. However, these methods encounter challenges such as the lack of coordination between different part motions and difficulties for networks to understand part concepts. Moreover, introducing finer-grained part concepts poses computational complexity challenges. In this paper, we propose Part-Coordinating Text-to-Motion Synthesis (ParCo), endowed with enhanced capabilities for understanding part motions and communication among different part motion generators, ensuring a coordinated and fined-grained motion synthesis. Specifically, we discretize whole-body motion into multiple part motions to establish the prior concept of different parts. Afterward, we employ multiple lightweight generators designed to synthesize different part motions and coordinate them through our part coordination module. Our approach demonstrates superior performance on common benchmarks with economic computations, including HumanML3D and KIT-ML, providing substantial evidence of its effectiveness. The code is available in the supplementary material.

Generating human motion from text has been dominated by denoising motion models either through diffusion or generative masking process. However, these models face great limitations in usability by requiring prior knowledge of the motion length. Conversely, autoregressive motion models address this limitation by adaptively predicting motion endpoints, at the cost of degraded generation quality and editing capabilities. To address these challenges, we propose Bidirectional Autoregressive Motion Model (BAMM), a novel text-to-motion generation framework. BAMM consists of two key components: (1) a motion tokenizer that transforms 3D human motion into discrete tokens in latent space, and (2) a masked self-attention transformer that autoregressively predicts randomly masked tokens via a hybrid attention masking strategy. By unifying generative masked modeling and autoregressive modeling, BAMM captures rich and bidirectional dependencies among motion tokens, while learning the probabilistic mapping from textual inputs to motion outputs with dynamically-adjusted motion sequence length. This feature enables BAMM to simultaneously achieving high-quality motion generation with enhanced usability and built-in motion editability. Extensive experiments on HumanML3D and KIT-ML datasets demonstrate that BAMM surpasses current state-of-the-art methods in both qualitative and quantitative measures.

Sign language videos are an important medium for spreading and learning sign language. However, most existing human image synthesis methods produce sign language images with details that are distorted, blurred, or structurally incorrect. They also produce sign language video frames with poor temporal consistency, with anomalies such as flickering and abrupt detail changes between the previous and next frames. To address these limitations, we propose a novel Pose-Guided Motion Model (PGMM) for generating fine-grained and motion-consistent sign language videos. Firstly, we propose a new Coarse Motion Module (CMM), which completes the deformation of features by optical flow warping, thus transfering the motion of coarse-grained structures without changing the appearance; Secondly, we propose a new Pose Fusion Module (PFM), which guides the modal fusion of RGB and pose features, thus completing the fine-grained generation. Finally, we design a new metric, Temporal Consistency Difference (TCD) to quantitatively assess the degree of temporal consistency of a video by comparing the difference between the frames of the reconstructed video and the previous and next frames of the target video. Extensive qualitative and quantitative experiments show that our method outperforms state-of-the-art methods in most benchmark tests, with visible improvements in details and temporal consistency.

We study the problem of generating intermediate images from image pairs with large motion while maintaining semantic consistency. Due to the large motion, the intermediate semantic information may be absent in input images. Existing methods either limit to small motion or focus on topologically similar objects, leading to artifacts and inconsistency in the interpolation results. To overcome this challenge, we delve into pre-trained image diffusion models for their capabilities in semantic cognition and representations, ensuring consistent expression of the absent intermediate semantic representations with the input. To this end, we propose DreamMover, a novel image interpolation framework with three main components: 1) A natural flow estimator based on the diffusion model that can implicitly reason about the semantic correspondence between two images. 2) To avoid the loss of detailed information during fusion, our key insight is to fuse information in two parts, high-level space and low-level space. 3) To enhance the consistency between the generated images and input, we propose the self-attention concatenation and replacement approach. Lastly, we present a challenging benchmark dataset called InterpBench to evaluate the semantic consistency of generated results. Extensive experiments demonstrate the effectiveness of our method. Code will be released soon.

In recent years, diffusion models have made remarkable strides in text-to-video generation, sparking a quest for enhanced control over video outputs to more accurately reflect user intentions. Traditional efforts predominantly focus on employing either semantic cues, like images or depth maps, or motion-based conditions, like moving sketches or object bounding boxes. Semantic inputs offer a rich scene context but lack detailed motion specificity; conversely, motion inputs provide precise trajectory information but miss the broader semantic narrative. For the first time, we integrate both semantic and motion cues within a diffusion model for video generation. To this end, we introduce the Scene and Motion Conditional Diffusion (SMCD), a novel methodology for managing multimodal inputs. It incorporates a recognized motion conditioning module~\cite{li2023gligen} and investigates various approaches to integrate scene conditions, promoting synergy between different modalities. For model training, we separate the conditions for the two modalities, introducing a two-stage training pipeline. Experimental results demonstrate that our design significantly enhances video quality, motion precision, and semantic coherence.

We introduce V-Trans4Style, an innovative algorithm tailored for dynamic video content editing needs. It is designed to adapt videos to different production styles like documentaries, dramas, feature films, or a specific YouTube channel's video-making technique. Our algorithm recommends optimal visual transitions to help achieve this flexibility using a more bottom-up approach. We first employ a transformer-based encoder-decoder network to learn recommending temporally consistent and visually seamless sequences of visual transitions using only the input videos. We then introduce a style conditioning module that leverages this model to iteratively adjust the visual transitions obtained from the decoder through activation maximization. We demonstrate the efficacy of our method through experiments conducted on our newly introduced AutoTransition++ dataset. It is a 6k video version of AutoTransition Dataset that additionally categorizes its videos into different production style categories. Our encoder-decoder model outperforms the state-of-the-art transition recommendation method, achieving improvements of 10% to 80% in Recall@K and mean rank values over baseline. Our style conditioning module results in visual transitions that improve the capture of the desired video production style characteristics by an average of around 12% in comparison to other methods when measured with similarity metrics. We hope that our work serves as a …

Video generation models have shown their superior ability to generate photo-realistic video. However, how to accurately control (or edit) the video remains a formidable challenge. The main issues are: 1) how to perform direct and accurate user control in editing; 2) how to execute editings like changing shape, expression, and layout without unsightly distortion and artifacts to the edited content; and 3) how to maintain spatio-temporal consistency of video after editing. To address the above issues, we propose DragVideo, a general drag-style video editing framework. Inspired by DragGAN, DragVideo addresses issues 1) and 2) by proposing the drag-style video latent optimization method which gives desired control by updating noisy video latent according to drag instructions through video-level drag objective function. We amend issue 3) by integrating the video diffusion model with sample-specific LoRA and Mutual Self-Attention in DragVideo to ensure the edited result is spatio-temporally consistent. We also present a series of testing examples for drag-style video editing and conduct extensive experiments across a wide array of challenging editing tasks, such as motion, skeleton editing, etc, underscoring DragVideo can edit video in an intuitive, faithful to the user's intention manner, with nearly unnoticeable distortion and artifacts, while maintaining spatio-temporal consistency. …

Story visualization aims to generate a series of realistic and coherent images based on a storyline. Current models adopt a frame-by-frame architecture by transforming the pre-trained text-to-image model into an auto-regressive manner. Although these models have shown notable progress, there are still three flaws. 1) The unidirectional generation of auto-regressive manner restricts the usability in many scenarios. 2) The additional introduced story history encoders bring an extremely high computational cost. 3) The story visualization and continuation models are trained and inferred independently, which is not user-friendly. To these ends, we propose a bidirectional, unified, and efficient framework, namely StoryImager. The StoryImager enhances the storyboard generative ability inherited from the pre-trained text-to-image model for a bidirectional generation. Specifically, we introduce a Target Frame Masking Strategy to extend and unify different story image generation tasks. Furthermore, we propose a Frame-Story Cross Attention Module that decomposes the cross attention for local fidelity and global coherence. Moreover, we design a Contextual Feature Extractor to extract contextual information from the whole storyline. The extensive experimental results demonstrate the excellent performance of our StoryImager.

The diffusion model is widely leveraged for either video generation or video editing. As each field has its task-specific problems, it is difficult to merely develop a single diffusion for completing both tasks simultaneously. Video diffusion sorely relying on the text prompt can be adapted to unify the two tasks. However, it lacks a high capability of aligning heterogeneous modalities between text and image, leading to various misalignment problems. In this work, we are the first to propose a unified Multi-alignment Diffusion, dubbed as MagDiff, for both tasks of high-fidelity video generation and editing. The proposed MagDiff introduces three types of alignments, including subject-driven alignment, adaptive prompts alignment, and high-fidelity alignment. Particularly, the subject-driven alignment is put forward to trade off the image and text prompts, serving as a unified foundation generative model for both tasks. The adaptive prompts alignment is introduced to emphasize different strengths of homogeneous and heterogeneous alignments by assigning different values of weights to the image and the text prompts. The high-fidelity alignment is developed to further enhance the fidelity of both video generation and editing by taking the subject image as an additional model input. Experimental results on four benchmarks suggest that our method outperforms …

Current image editing methods primarily utilize DDIM Inversion, employing a two-branch diffusion approach to preserve the attributes and layout of the original image. However, these methods encounter challenges with non-rigid edits, which involve altering the image's layout or structure. Our comprehensive analysis reveals that the high-frequency components of DDIM latent, crucial for retaining the original image's key features and layout, significantly contribute to these limitations. Addressing this, we introduce FlexiEdit, which enhances fidelity to input text prompts by refining DDIM latent, specifically by reducing high-frequency components in targeted editing areas. FlexiEdit comprises two key components: (1) Latent Refinement, which modifies DDIM latent to better accommodate layout adjustments, and (2) Edit Fidelity Enhancement via Re-inversion, aimed at ensuring the edits more accurately reflect the input text prompts. Our approach represents notable progress in image editing, particularly in performing complex non-rigid edits, showcasing its enhanced capability through comparative experiments.

We introduce a novel diffusion transformer, LazyDiffusion, that generates partial image updates efficiently, targeting interactive image editing applications. Starting from a blank canvas or an image, a user specifies a sequence of localized image modifications using a binary mask and a text prompt. Our generator operates in two phases. First, a context encoder processes the current canvas and user mask to produce a compact global context tailored to the region to generate. Second, conditioned on this global context, a diffusion-based decoder synthesizes the masked pixels in a ``lazy'' fashion, i.e., it only generates the masked region. This contrasts with previous works that either regenerate the full canvas, wasting time and computation, or confine processing to a tight rectangular crop around the mask, ignoring the global image context altogether. Our decoder's runtime and computation cost scale with the mask size, which is typically small for interactive edits. Since the diffusion process dominates the runtime and cost, our encoder introduces negligible overhead. Our approach amortizes the generation cost over several user interactions, making the editing experience more interactive, especially at high-resolution. We train LazyDiffusion on a large scale text-to-image dataset at 1024x1024 resolution. We demonstrate that our approach is competitive with state-of-the-art …

While style transfer techniques have been well-developed for 2D image stylization, the extension of these methods to 3D scenes remains relatively unexplored. Existing approaches demonstrate proficiency in transferring colors and textures but often struggle with replicating the geometry of the scenes. In our work, we leverage an explicit Gaussian Scale (GS) representation and directly match the distributions of Gaussians between style and content scenes using the Earth Mover's Distance (EMD). By employing the entropy-regularized Wasserstein-2 distance, we ensure that the transformation maintains spatial smoothness. Additionally, we decompose the scene stylization problem into smaller chunks to enhance efficiency. This paradigm shift reframes stylization from a pure generative process driven by latent space losses to an explicit matching of distributions between two Gaussian representations. Our method achieves high-resolution 3D stylization by faithfully transferring details from 3D style scenes onto the content scene. Furthermore, WaSt-3D consistently delivers results across diverse content and style scenes without necessitating any training, as it relies solely on optimization-based techniques.

This paper introduces innovative solutions to enhance spatial controllability in diffusion models reliant on text queries. We first introduce vision guidance as a foundational spatial cue within the perturbed distribution. This significantly refines the search space in a zero-shot paradigm to focus on the image sampling process adhering to the spatial layout conditions. To precisely control the spatial layouts of multiple visual concepts with the employment of vision guidance, we propose a universal framework, Layered Rendering Diffusion (LRDiff), which constructs an image-rendering process with multiple layers, each of which applies the vision guidance to instructively estimate the denoising direction for a single object. Such a layered rendering strategy effectively prevents issues like unintended conceptual blending or mismatches while allowing for more coherent and contextually accurate image synthesis. The proposed method offers a more efficient and accurate means of synthesising images that align with specific layout and contextual requirements. Through experiments, we demonstrate that our method outperforms existing techniques, both quantitatively and qualitatively, in two specific layout-to-image tasks: bounding box-to-image and instance mask-to-image. Furthermore, we extend the proposed framework to enable spatially controllable editing.

Images tell stories, trigger emotions, and let us recall memories -- they make us think. Thus, they have the ability to attract and hold one's attention, which is the definition of being ``interesting''. Yet, the appeal of an image is highly subjective. Looking at the image of my boy taking his first steps will always bring me back to this emotional moment, while it is just a blurry, quickly taken snapshot to most others. Preferences vary widely: some adore cats, others are dog enthusiasts, and a third group may not be fond of either. We argue that every image can be interesting to a particular observer under certain circumstances. This work particularly emphasizes subjective preferences. However, our analysis of 2.5k image collections from diverse users of the photo-sharing platform Flickr reveals that specific image characteristics make them commonly more interesting. For instance, images, including professionally taken landscapes, appeal broadly due to their aesthetic qualities. In contrast, subjectively interesting images, such as those depicting personal or niche community events, resonate on a more individual level, often evoking personal memories and emotions.

Recent advances in image editing have been driven by the development of denoising diffusion models, marking a significant leap forward in this field. Despite these advances, the generalization capabilities of recent image editing approaches remain constrained. In response to this challenge, our study introduces a novel image editing framework with enhanced generalization robustness by boosting in-context learning capability and unifying language instruction. This framework incorporates a module specifically optimized for image editing tasks, leveraging the VMamba Block and an editing-shift matching strategy to augment in-context learning. Furthermore, we unveil a selective area-matching technique specifically engineered to address and rectify corrupted details in generated images, such as human facial features, to further improve the quality. Another key innovation of our approach is the integration of a language unification technique, which aligns language embeddings with editing semantics to elevate the quality of image editing. Moreover, we compile the first dataset for image editing with visual prompts and editing instructions that could be used to enhance in-context capability. Trained on this dataset, our methodology not only achieves superior synthesis quality for trained tasks, but also demonstrates robust generalization capability across unseen vision tasks through tailored prompts.

Text-to-image diffusion models allow users control over the content of generated images. Still, text-to-image generation occasionally leads to generation failure requiring users to generate dozens of images under the same text prompt before they obtain a satisfying result. We formulate the lottery ticket hypothesis in denoising: randomly initialized Gaussian noise images contain special pixel blocks (winning tickets) that naturally tend to be denoised into specific content independently. The generation failure in standard text-to-image synthesis is caused by the gap between optimal and actual spatial distribution of winning tickets in initial noisy images. To this end, we implement semantic-driven initial image construction creating initial noise from known winning tickets for each concept mentioned in the prompt. We conduct a series of experiments that verify the properties of winning tickets and demonstrate that the winning tickets have generalizability across images and prompts. Our results show that aggregated winning tickets effectively induce the model to spontaneously generate the object at the corresponding location.

Text-to-image generative models have become a prominent and powerful tool that excels at generating high-resolution realistic images. However, guiding the generative process of these models to take into account detailed forms of conditioning reflecting style and/or structure information remains an open problem. In this paper, we present LoRAdapter, an approach that unifies both style and structure conditioning under the same formulation using a novel conditional LoRA block that enables zero-shot control. LoRAdapter is an efficient and powerful approach to condition text-to-image diffusion models, which enables fine-grained control conditioning during generation and outperforms recent state-of-the-art approaches.

Text-based image editing and generation methods have diverse real-world applications. However, text-guided infinite image synthesis faces several challenges. First, there is a lack of text-image paired datasets with high-resolution and contextual diversity. Second, expanding images based on text requires global coherence and rich local context understanding. Previous studies have mainly focused on limited categories, such as natural landscapes, and also required to train on high-resolution images with paired text. To address these challenges, we propose a novel approach utilizing Large Language Models (LLMs) for both global coherence and local context understanding, without any high-resolution text-image paired training dataset. We train the diffusion model to expand an image conditioned on global and local captions generated from the LLM and visual feature. At the inference stage, given an image and a global caption, we use the LLM to generate a next local caption to expand the input image. Then, we expand the image using the global caption, generated local caption and the visual feature to consider global consistency and spatial local context. In experiments, our model outperforms the baselines both quantitatively and qualitatively. Furthermore, our model demonstrates the capability of text-guided arbitrary-sized image generation in zero-shot manner with LLM guidance.

Recent years have witnessed the success of large text-to-image diffusion models and their remarkable potential to generate high-quality images. The further pursuit of enhancing the editability of images has sparked significant interest in the downstream task of inpainting a novel object described by a text prompt within a designated region in the image. Nevertheless, the problem is not trivial from two aspects: 1) Solely relying on one single U-Net to align text prompt and visual object across all the denoising timesteps is insufficient to generate desired objects; 2) The controllability of object generation is not guaranteed in the intricate sampling space of diffusion model. In this paper, we propose to decompose the typical single-stage object inpainting into two cascaded processes: 1) semantic pre-inpainting that infers the semantic features of desired objects in a multi-modal feature space; 2) high-fieldity object generation in diffusion latent space that pivots on such inpainted semantic features. To achieve this, we cascade a Transformer-based semantic inpainter and an object inpainting diffusion model, leading to a novel CAscaded Transformer-Diffusion (CAT-Diffusion) framework for text-guided object inpainting. Technically, the semantic inpainter is trained to predict the semantic features of the target object conditioning on unmasked context and text prompt. …

Customized generation aims to incorporate a novel concept into a pre-trained text-to-image model, enabling new generations of the concept in novel contexts guided by textual prompts. However, customized generation suffers from an inherent trade-off between concept fidelity and editability, i.e., between precisely modeling the concept and faithfully adhering to the prompts. Previous methods reluctantly seek a compromise and struggle to achieve both high concept fidelity and ideal prompt alignment simultaneously. In this paper, we propose a "Divide, Conquer, then Integrate" (DCI) framework, which performs a surgical adjustment in the early stage of denoising to liberate the fine-tuned model from the fidelity-editability trade-off at inference.The two conflicting components in the trade-off are decoupled and individually conquered by two collaborative branches, which are then selectively integrated to preserve high concept fidelity while achieving faithful prompt adherence. To obtain a better fine-tuned model, we introduce an Image-specific Context Optimization (ICO) strategy for model customization. ICO replaces manual prompt templates with learnable image-specific contexts, providing an adaptive and precise fine-tuning direction to promote the overall performance. Extensive experiments demonstrate the effectiveness of our method in reconciling the fidelity-editability trade-off, particularly for generations with weak model priors. Code has made anonymously available at https://anonymous.4open.science/r/DCI_ICO-90C2

Text-to-Image (T2I) generation has made significant advancements with the advent of diffusion models. These models exhibit remarkable ability to produce images based on textual prompts. Current T2I models allow users to specify object colors using linguistic color names. However, these labels encompass broad color ranges, making it difficult to achieve precise color matching. To tackle this challenging task, named as color prompt learning, we propose to learn specific color prompts tailored to user-selected colors. These prompts are finally employed to generate objects with the exact desired colors. Observing the existing T2I adaptation approaches cannot achieve satisfactory performance, we propose to generate basic geometric objects in the target color. Leveraging color and shape disentanglement, our method, denoted as ColorPeel, successfully assists the T2I models to peel off the novel color prompts from these colored shapes. In the experiments, we demonstrate the efficacy of ColorPeel in achieving precise color generation with T2I models and generalize ColorPeel to effectively learn abstract attribute concepts, including textures, materials, etc. Our findings provide a valuable step towards improving precision and versatility of T2I models, offering new opportunities for creative applications and design tasks.

Personalized image generation involves creating images aligned with an individual’s visual preference. Current generative models are, however, tuned to produce outputs that appeal to a broad audience, and personalization to individual users' visual preferences relies on iterative and manual prompt engineering by the user, which is neither time-efficient nor scalable. We propose to personalize the image generation process by first inviting users to comment on a small selection of images, explaining why they like or dislike each. Based on these comments, we infer a user’s liked and disliked visual attributes, i.e., their visual preference, using a large language model. These attributes are used to guide a text-to-image model toward producing images that are personalized towards the individual user's visual preference. Through a series of user tests and large language model guided evaluations, we demonstrate that our proposed method results in generations that are well aligned with individual users' visual preferences.

The deployment of large-scale text-to-image diffusion models on mobile devices is impeded by their substantial model size and high latency. In this paper, we present \textbf{MobileDiffusion}, an ultra-efficient text-to-image diffusion model obtained through extensive optimizations in both architecture and sampling techniques. We conduct a comprehensive examination of model architecture design to minimize model size and FLOPs, while preserving image generation quality. Additionally, we revisit the advanced sampling technique by diffusion-GAN, and make one-step sampling compatible to downstream applications trained on the base model. Empirical studies, conducted both quantitatively and qualitatively, demonstrate the effectiveness of our proposed technologies. With them, MobileDiffusion achieves instant text-to-image generation on mobile devices, establishing a new state of the art.

Text-to-image (T2I) diffusion models have shown significant success in personalized text-to-image generation, which aims to generate novel images with human identities indicated by the reference images. Despite promising identity fidelity has been achieved by several tuning-free methods, they usually suffer from overfitting issues. The learned identity tends to entangle with irrelevant information, resulting in unsatisfied text controllability, especially on faces. In this work, we present MasterWeaver, a test-time tuning-free method designed to generate personalized images with both faithful identity fidelity and flexible editability. Specifically, MasterWeaver adopts an encoder to extract identity features and steers the image generation through additional introduced cross attention. To improve editability while maintaining identity fidelity, we propose an editing direction loss for training, which aligns the editing directions of our MasterWeaver with those of the original T2I model. Additionally, a face-augmented dataset is constructed to facilitate disentangled identity learning, and further improve the editability. Extensive experiments demonstrate that our MasterWeaver can not only generate personalized images with faithful identity, but also exhibit superiority in text controllability. Our code will be made publicly available.

In the e-commerce realm, compelling advertising images are pivotal for attracting customer attention. While generative models automate image generation, they often produce substandard images that may mislead customers and require significant labor costs to inspect. This paper delves into increasing the rate of available generated images. We first introduce a multi-modal Reliable Feedback Network (RFNet) to automatically inspect the generated images. Combining the RFNet into a recurrent process, Recurrent Generation, results in a higher number of available advertising images. To further enhance production efficiency, we fine-tune diffusion models with an innovative Consistent Condition regularization utilizing the feedback from RFNet (RFFT). This results in a remarkable increase in the available rate of generated images, reducing the number of attempts in Recurrent Generation, and providing a highly efficient production process without sacrificing visual appeal. We also construct a Reliable Feedback 1 Million (RF1M) dataset which comprises over one million generated advertising images annotated by humans, which helps to train RFNet to accurately assess the availability of generated images and faithfully reflect the human feedback. Generally speaking, our approach offers a reliable solution for advertising image generation. The dataset and code will be released after acceptance.

Advancements in open-sourced text-to-image models and fine-tuning methods have led to the increasing risk of malicious adaptation, i.e., fine-tuning to generate harmful/unauthorized content. Recent works, e.g., Glaze or MIST, have developed data-poisoning techniques which protect the data against adaptation methods. In this work, we consider an alternative paradigm for protection. We propose to ``immunize'' the model by learning model parameters that are difficult for the adaptation methods when fine-tuning malicious content; in short IMMA. Specifically, IMMA should be applied before the release of the model weights to mitigate these risks. Empirical results show IMMA's effectiveness against malicious adaptations, including mimicking the artistic style and learning of inappropriate/unauthorized content, over three adaptation methods: LoRA, Textual-Inversion, and DreamBooth.

While large text-to-image diffusion models have made significant progress in high-quality image generation, challenges persist when users insert their portraits into existing photos, especially group photos. Concretely, existing customization methods struggle to insert facial identities at desired locations in existing images, and it is difficult for existing local image editing methods to deal with facial details. To address these limitations, we propose AddMe, a powerful diffusion-based portrait generator that can insert a given portrait into a desired location in an existing scene image in a zero-shot manner. Specifically, we propose a novel identity adapter to learn a facial representation decoupled from existing characters in the scene. Meanwhile, to ensure that the generated portrait can interact properly with others in the existing scene, we design an enhanced portrait attention module to capture contextual information during the generation process. Our method is compatible with both text and various spatial conditions, enabling precise control over the generated portraits. Extensive experiments demonstrate significant improvements in both performance and efficiency.

Image processing, including image restoration, image enhancement, etc., involves generating a high-quality clean image from a degraded input. Deep learning-based methods have shown superior performance for various image processing tasks in terms of single-task conditions. However, they require to train separate models for different degradations and levels, which limits the generalization abilities of these models and restricts their applications in real-world. In this paper, we propose a text-induced Unified image Processor for low-level vision tasks, termed UniProcessor, which can effectively process various degradation types and levels, and support multimodal control. Specifically, our UniProcessor encodes degradation-specific information with the subject prompt and process degradations with the manipulation prompt. These context control features are injected into the UniProcessor backbone via cross-attention to control the processing procedure. For automatic subject-prompt generation, we further build a vision-language model for general-purpose low-level degradation perception via instruction tuning techniques. Our UniProcessor covers 30 degradation types, and extensive experiments demonstrate that our UniProcessor can well process these degradations without additional training or tuning and outperforms other competing methods. Moreover, with the help of degradation-aware context control, our UniProcessor first shows the ability to individually handle a single distortion in an image with multiple degradations.

Diffusion models, known for their tremendous ability to generate novel and high-quality samples, have recently raised concerns due to their data memorization behavior, which poses privacy risks. Recent approaches for memory mitigation either only focused on the text modality problem in cross-modal generation tasks or utilized data augmentation strategies. In this paper, we propose a novel training framework for Diffusion models from the perspective of visual modality, which is more generic and fundamental for mitigating memorization. To facilitate ``forgetting'' of stored information in Diffusion model parameters, we propose an iterative ensemble training strategy by splitting the data into multiple shards for training multiple models and intermittently aggregating these model parameters. Moreover, practical analysis on losses illustrates that the training loss for easily memorable images tends to be obviously lower. Thus, we propose an anti-gradient control method to exclude the sample with a lower loss value from the current mini-batch to avoid memorizing. Extensive experiments and analysis on three datasets are conducted to illustrate the effectiveness of our method, and results show that our method successfully reduces memory capacity while even improving the performance slightly. Moreover, to save the computing cost, we successfully apply our method to fine-tune the well-trained Diffusion …