Reproduce “Real-world Image Dehazing with Coherence-based Pseudo Labeling and Cooperative Unfolding Network” 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

Chengyu Fang, Chunming He, Fengyang Xiao, Yulun Zhang, Longxiang Tang, Yuelin Zhang, Kai Li, Xiu Li

This research introduces a Cooperative Unfolding Network (CORUN) and a Coherence-based Pseudo Labeling (Colabator) framework to tackle real-world image dehazing challenges. This integrated system achieves superior performance on the RTTS dataset, with the lowest FADE (0.824) and BRISQUE (11.956) scores, highest NIMA (5.342), and notably enhances object detection in dehazed images.

Key insights

  • Dehazing performance in real-world conditions improves when physical models (like ASM) are deeply integrated into a neural network architecture, offering better interpretability and detail preservation.
  • Generating high-quality pseudo-labels for unlabeled real-world data through a robust iterative mean-teacher framework and regional quality assessment (Colabator) is crucial for overcoming data scarcity and enhancing model generalization.
  • Combining a global coherence loss with density-aware semantic losses during training reinforces physical consistency and improves the visual quality of dehazed images.

Summary by alphaXiv.

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