Abstract:Supervised dehazing models, trained on synthetic hazy-clean image pairs, often face a notable decline in performance when applied to real-world scenes. Consequently, CycleGAN-based unpaired dehazing methods are proposed to improve the model’s generalization. One successful approach among these methods involves decomposing the physical properties of the atmospheric scattering model (ASM). However, estimating physical properties individually from input images is difficult without supervised labels, which ignores the semantic consistency between different physical regions. We claim semantic region information can offer additional geometric spatial constraints for estimating physical properties, as natural images can be divided into regions with similar scene depths. Motivated by this, we propose a novel generalized and realistic unpaired image dehazing framework via region-aware physical constraints (RPC-Dehaze). Our approach utilizes fine-grained semantic region maps from the Segment Anything Model (SAM) in a specially designed region prompt enhancement module. This enables the dehazing and hazing cyclic networks to learn region-aware physical constraints, leading to accurate estimation of haze imaging physical properties. In contrast to existing unpaired methods that treat dehazing and hazing networks equally, we incorporate Retinex theory into the hazing network, allowing it to learn diverse illumination effects in different regions. We adaptively refine the Retinex-based illumination component, resulting in more realistic hazy images. To further facilitate unsupervised learning in our framework, we propose a physical consensual contrastive regularization to ensure compact representation constraints in the latent feature space. Extensive experiments on synthetic and real image datasets show our method surpasses state-of-the-art unpaired dehazing methods in both effectiveness and generalization capability.

Robust Real-World Image Dehazing Via Knowledge Guided Conditional Diffusion Model Finetuning

From Coarse to Fine: A Two-Stage Network for Dense Haze removal

Frequency compensated diffusion model for real-scene dehazing

Advancing Real-World Image Dehazing: Perspective, Modules, and Training

Remote Sensing Image Dehazing Using Adaptive Region-Based Diffusion Models

SCL-Dehaze: Toward Real-World Image Dehazing via Semi-Supervised Codebook Learning

Accurate and lightweight dehazing via multi-receptive-field non-local network and novel contrastive regularization

RSHazeDiff: A Unified Fourier-aware Diffusion Model for Remote Sensing Image Dehazing

Contrastive adaptive frequency decomposition network guided by haze discrimination for real-world image dehazing

From Synthetic to Real: Image Dehazing Collaborating with Unlabeled Real Data

Through the Real World Haze Scenes: Navigating the Synthetic-to-Real Gap in Challenging Image Dehazing

Toward Generalized and Realistic Unpaired Image Dehazing Via Region-Aware Physical Constraints

High-quality Image Dehazing with Diffusion Model

Hierarchical Integration Diffusion Model for Realistic Image Deblurring

Generative Adversarial and Self-Supervised Dehazing Network

GridDehazeNet+: An Enhanced Multi-Scale Network with Intra-Task Knowledge Transfer for Single Image Dehazing

Haze-robust Image Understanding Via Context-Aware Deep Feature Refinement.

Single image mixed dehazing method based on numerical iterative model and DehazeNet

Non-aligned supervision for Real Image Dehazing