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

TL;DR

This work tackles real-world image dehazing by introducing CORUN, a cooperative unfolding network that jointly optimizes atmospheric scattering and scene content within a physics-informed framework. To address data scarcity and domain gaps, it also proposes Colabator, an iterative mean-teacher system that generates high-quality pseudo labels using a dynamic, coherence-based selection and an optimal label pool. The combination yields state-of-the-art RID performance on real datasets and demonstrates strong generalization when Colabator is applied to other dehazing methods. The approach enhances dehazing quality, color fidelity, and downstream task performance, with code forthcoming for reproducibility.

Abstract

Real-world Image Dehazing (RID) aims to alleviate haze-induced degradation in real-world settings. This task remains challenging due to the complexities in accurately modeling real haze distributions and the scarcity of paired real-world data. To address these challenges, we first introduce a cooperative unfolding network that jointly models atmospheric scattering and image scenes, effectively integrating physical knowledge into deep networks to restore haze-contaminated details. Additionally, we propose the first RID-oriented iterative mean-teacher framework, termed the Coherence-based Label Generator, to generate high-quality pseudo labels for network training. Specifically, we provide an optimal label pool to store the best pseudo-labels during network training, leveraging both global and local coherence to select high-quality candidates and assign weights to prioritize haze-free regions. We verify the effectiveness of our method, with experiments demonstrating that it achieves state-of-the-art performance on RID tasks. Code will be available at https://github.com/cnyvfang/CORUN-Colabator.

Figures (8)

• Figure 1: Results of cutting-edge methods. Our CORUN better restores hazy-contaminated details. Furthermore, techniques optimized by our Colabator framework, indicated by a "+" suffix, exhibit strong generalization in haze removal and color correction.
• Figure 2: The architecture of the proposed CORUN with the details at $k^{th}$ stage.
• Figure 3: The plug-and-play Coherence-based Pseudo-label Generator paradigm.
• Figure 4: Visual comparison on RTTSli2019benchmarking. Please zoom in for a better view.
• Figure 5: Visual comparison on Fattal’s datafattal2014dehazing.