Reproduce “Gated Fusion Network for Single Image Dehazing” 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

Wenqi Ren, Lin Ma, Jiawei Zhang, Jinshan Pan, Xiaochun Cao, Wei Liu, Ming-Hsuan Yang

Researchers from Chinese Academy of Sciences, Tencent AI Lab, and UC Merced developed a Gated Fusion Network for single image dehazing that directly restores clear images by learning to fuse multiple pre-processed inputs. This end-to-end model surpasses existing state-of-the-art methods in quantitative metrics and visual quality, effectively removing haze and artifacts without explicit estimation of atmospheric light or transmission.

Key insights

  • Directly learning to fuse multiple pre-processed image inputs via learned pixel-wise confidence maps offers a robust way to bypass the ill-posed problem of estimating transmission and atmospheric light.
  • A multi-scale network architecture is crucial for effectively mitigating halo artifacts and refining details in dehazing results.
  • Combining content loss at multiple scales with an adversarial loss helps produce both pixel-accurate and visually realistic dehazed images.

Summary by alphaXiv.

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