What to Remove, What to Preserve: Dual-Ambiguity Rectification for All-in-One Image Restoration
DAR-Net separates degradation cues from scene content to reduce both leftover artifacts and content damage.
The paper frames all-in-one image restoration as a “dual ambiguity” problem: models can confuse what should be removed with what should be preserved. Its proposed DAR-Net uses archetype mixture modeling to structure degradation states, then applies separate semantic and spatial rectification modules. In reported benchmarks, it leads three- and five-degradation settings, with average PSNR gains of 0.14 dB and 0.34 dB over the strongest competitor. ArXiv · AI/CL/LG's note
The paper frames all-in-one image restoration as a “dual ambiguity” problem: models can confuse what should be removed with what should be preserved. Its proposed DAR-Net uses archetype mixture modeling to structure degradation states, then applies separate semantic and spatial rectification modules. In reported benchmarks, it leads three- and five-degradation settings, with average PSNR gains of 0.14 dB and 0.34 dB over the strongest competitor. ArXiv · AI/CL/LG's note
score 4