New ExpoMotion benchmark and HOP network tackle multi-exposure fusion challenges

Researchers have introduced ExpoMotion, a large-scale benchmark dataset designed to evaluate multi-exposure fusion (MEF) deghosting capabilities in dynamic scenes. The dataset includes 1,738 sequences and 10,909 images, addressing the limitations of existing benchmarks that often neglect motion and lack reliable ground truth. To handle the complexities within ExpoMotion, a new Householder Orthogonal Projection (HOP) network is proposed. This network decouples alignment into exposure pre-alignment and ghost filtering, utilizing a Global Priors Illumination Alignment module for exposure harmonization and a Householder Orthogonal Attention module to project artifacts out of the feature manifold. AI

IMPACT Introduces a new benchmark and model for improving image quality in dynamic, multi-exposure photography.

RANK_REASON The cluster contains a research paper introducing a new benchmark dataset and a novel network architecture for a specific computer vision task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• ExpoMotion
• Global Priors Illumination Alignment
• Householder Orthogonal Attention
• Householder Orthogonal Projection network

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →