Abstract:Recently, Unsupervised algorithms has achieved remarkable performance in image dehazing. However, the CycleGAN framework can lead to confusion in generator learning due to inconsistent data distributions, and the DisentGAN framework lacks effective constraints on generated images, resulting in the loss of image content details and color distortion. Moreover, Squeeze and Excitation channel attention employs only fully connected layers to capture global information, lacking interaction with local information, resulting in inaccurate feature weight allocation for image dehazing. To solve the above problems, in this paper, we propose an Unsupervised Bidirectional Contrastive Reconstruction and Adaptive Fine-Grained Channel Attention Networks (UBRFC-Net). Specifically, an Unsupervised Bidirectional Contrastive Reconstruction Framework (BCRF) is proposed, aiming to establish bidirectional contrastive reconstruction constraints, not only to avoid the generator learning confusion in CycleGAN but also to enhance the constraint capability for clear images and the reconstruction ability of the unsupervised dehazing network. Furthermore, an Adaptive Fine-Grained Channel Attention (FCA) is developed to utilize the correlation matrix to capture the correlation between global and local information at various granularities promotes interaction between them, achieving more efficient feature weight assignment. Experimental results on challenging benchmark datasets demonstrate the superiority of our UBRFC-Net over state-of-the-art unsupervised image dehazing methods. This study successfully introduces an enhanced unsupervised image dehazing approach, addressing limitations of existing methods and achieving superior dehazing results. The source code is available at https://github.com/Lose-Code/UBRFC-Net.

Addressing Domain Discrepancy: A Dual-branch Collaborative Model to Unsupervised Dehazing

DADRnet: Cross-domain Image Dehazing Via Domain Adaptation and Disentangled Representation.

MsDA: Multi-scale Domain Adaptation Dehazing Network

Depth-agnostic Single Image Dehazing

Semi-Supervised Domain Alignment Learning for Single Image Dehazing

A Dual-Branch Residual Network for Inhomogeneous Dehazing

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

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

CrossDehaze: Scaling Up Image Dehazing with Cross-Data Vision Alignment and Augmentation

Unsupervised Bidirectional Contrastive Reconstruction and Adaptive Fine-Grained Channel Attention Networks for image dehazing

Dual-branch feature fusion dehazing network via multispectral channel attention

Depth Information Assisted Collaborative Mutual Promotion Network for Single Image Dehazing

Robust Unpaired Image Dehazing Via Density and Depth Decomposition

Unpaired Deep Image Dehazing Using Contrastive Disentanglement Learning

Multiscale Depth Fusion With Contextual Hybrid Enhancement Network for Image Dehazing

Semi-supervised Single-Image Dehazing Network via Disentangled Meta-Knowledge

Discrete Haze Level Dehazing Network

Two-step image dehazing with intra-domain and inter-domain adaptation

Non-aligned supervision for Real Image Dehazing

A Multi-Scale Dehazing Network with Dark Channel Priors

DCNet - Dual-Task Cycle Network for End-to-End Image Dehazing.