
Motivation & Challenge: Unpaired dehazing needs reliable restoration without costly diffusion sampling. Dehaze-RL learns physical parameters, enhancement factors, and spatial gating, guided by perceptual rewards.
Results & Venues: Improved restoration quality with a physics-grounded hybrid action space and multi-granularity rewards.
Accepted at ECCV 2026.
Link: https://media.eventhosts.cc/Conferences/ECCV2026/pdfs/3070.pdf
1 like · 0 views · Sep 15, 2026
Motivation & Challenge: Unpaired dehazing needs reliable restoration without costly diffusion sampling.