Reproduce “ClearAIR: A Human-Visual-Perception-Inspired All-in-One Image Restoration” 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

Xu Zhang, Huan Zhang, Guoli Wang, Qian Zhang, Lefei Zhang

ClearAIR introduces a human-visual-perception-inspired All-in-One Image Restoration (AiOIR) framework that addresses spatially non-uniform image degradations through a hierarchical, coarse-to-fine strategy. The framework consistently achieved superior performance across diverse synthetic and real-world degradation scenarios, including an average 0.22 dB PSNR gain over Histoformer in all-weather tasks and 0.62 dB PSNR over OneRestore on the CDD-11 dataset.

Key insights

  • Mimicking Human Visual Perception's (HVP) hierarchical, coarse-to-fine processing strategy effectively addresses spatially non-uniform degradations and complex textures.
  • Integrating global image quality assessment with region-aware semantic guidance and local degradation type recognition enhances the model's adaptability.
  • A self-supervised Internal Clue Reuse Mechanism (ICRM) can effectively leverage intrinsic image information to recover fine details and prevent oversmoothing.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.