ExpoMotion: A Large-Scale Benchmark and A Householder Projection Network for Multi-Exposure Fusion

Multi-Exposure Fusion (MEF) effectively extends dynamicrange, but practical deployment is hindered by motion-induced ghostingand the scarcity of high-quality dynamic benchmarks. Current bench-marks largely neglect dynamic scenes and lack reliable ground truth,making it difficult to handle the complexity of real-world motions. In re-sponse, we introduce ExpoMotion, a large-scale benchmark designed toevaluate deghosting capabilities. Comprising 1,738 sequences and 10,909images across diverse environments, it covers a wide range of motionsand provides high-fidelity GTs constructed through an expert-guided ac-quisition pipeline. To tackle the complex dynamics and extreme con-ditions captured in this benchmark, we propose the Householder Or-thogonal Projection network (HOP), which revisits MEF deghostingfrom a mathematical perspective via Householder transformation, de-coupling multi-frame alignment into exposure pre-alignment and ghostfiltering. Specifically, the Global Priors Illumination Alignment (GPIA)module first rectifies drastic dynamic range discrepancies by utilizingglobal statistics for exposure harmonization. Regarding ghost removal,our Householder Orthogonal Attention (HOA) models artifacts as or-thogonal perturbations. By employing a dynamic Householder reflector,HOA effectively projects ghosts out of the feature manifold while pre-serving high-frequency details. Experiments demonstrate that our Ex-poMotion dataset enables superior generalization and artifact-free detailrestoration, while also validating the effectiveness and efficiency of theHOP method. The dataset and code are available at https://github.com/Leo-LiuYao/ExpoMotion.