Reproduce “All-in-One Video Restoration under Smoothly Evolving Unknown Weather Degradations” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wenrui Li, Hongtao Chen, Yao Xiao, Wangmeng Zuo, Jiantao Zhou, Yonghong Tian, Xiaopeng Fan
ORCANet, an all-in-one recurrent conditional and adaptive prompting network, addresses video restoration under smoothly evolving unknown weather degradations (SEUD) by integrating physics-guided dehazing and a two-level flow-prompting mechanism. It achieved superior performance on a new benchmark, reaching up to 37.88 dB PSNR on MOT17 for single-type degradations and demonstrating enhanced robustness for compound degradation scenarios.
Key insights
- Explicitly modeling the smooth temporal evolution of degradation types and intensities, rather than frame-independent application, is crucial for realistic all-in-one video restoration.
- A two-level prompting mechanism, combining segment-level static prompts for degradation type and frame-level dynamic prompts for intensity variation, effectively captures both coarse degradation identity and fine-grained temporal changes.
- Physics-guided coarse intensity estimation and dehazing provide a robust initialization, significantly improving overall restoration performance and temporal stability, particularly for haze conditions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.