DDTNet: Degradation Disentanglement and Transfer Network for Test-Time All-in-One De-weathering Adaptation

2026-06-15 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

DDTNet, or Degradation Disentanglement and Transfer Network, introduces a novel approach to all-in-one adverse weather image restoration, aiming to overcome the performance compromises and domain gap issues prevalent in existing unified models. Instead of directly recovering clean content, DDTNet focuses on degradation transfer. It disentangles degradation patterns from target-domain degraded images and applies them to source-domain clean images, thereby generating domain-adaptive paired training data. This synthetic data is then used to fine-tune existing restoration models, significantly enhancing their adaptability across diverse weather conditions like rain, haze, and snow, and different domains. At its core, DDTNet employs a Degradation Disentanglement Module (DDM) with Degradation Coupled Attention (DCA) to effectively capture both general and weather-specific degradation features. Experimental evaluations confirm that DDTNet consistently and significantly improves the performance of current all-in-one models across real-world deraining, desnowing, and dehazing datasets.

Key takeaway

For computer vision engineers developing all-in-one de-weathering solutions, DDTNet offers a robust strategy to overcome performance compromises and domain adaptation challenges. You should consider integrating DDTNet's degradation disentanglement and transfer mechanism to generate high-quality, domain-adaptive training data. This approach allows you to fine-tune existing models effectively, significantly improving their real-world performance across diverse adverse weather conditions like rain, snow, and haze.

Key insights

DDTNet enhances all-in-one de-weathering by disentangling and transferring degradation patterns to generate domain-adaptive training data.

Principles

• Modeling degradation patterns is more feasible than recovering clean content.
• Disentangling degradation enables effective transfer.
• Domain-adaptive paired data improves model fine-tuning.

Method

DDTNet disentangles degradation patterns from target-domain degraded images, transfers them to source-domain clean images to create domain-adaptive paired data, then fine-tunes restoration models using these pairs.

In practice

• Apply DDTNet to fine-tune existing de-weathering models.
• Generate synthetic training data for domain adaptation.
• Improve performance on real-world deraining, desnowing, dehazing.

Topics

• All-in-One Image Restoration
• Degradation Disentanglement
• Domain Adaptation
• Test-Time Adaptation
• De-weathering
• Computer Vision

Best for: Research Scientist, AI Scientist, Computer Vision Engineer

Related on AIssential

• Adapting Diffusion Language Models for Lossless Pixel-Level Image Transmission — DDM-SSCC adapts diffusion language models for lossless pixel-level image transmission, achieving superior exact recovery via novel coding and denoising strategies.
• ErA: Error-Aware Deep Unrolling Network for Single Image Defocus Deblurring — ErA employs an error-aware deep unrolling network for state-of-the-art single-image defocus deblurring.
• Multi-Branch Non-Homogeneous Image Dehazing via Concentration Partitioning and Image Fusion — Decomposing non-homogeneous haze into homogeneous sub-problems enables targeted, multi-branch image enhancement and fusion.
• Multi-Modal Guided Multi-Source Domain Adaptation for Object Detection — Multi-modal inputs like depth and text can enhance multi-source domain adaptation for robust object detection.
• ATT-CR: Adaptive Triangular Transformer for Cloud Removal — Adaptive Triangular Transformer (ATT-CR) efficiently removes clouds from remote sensing images by combining linear triangular attention and feature-selective gating.
• Fusing Transferred Priors and Physics-based Decomposition for Underwater Image Enhancement — A novel UIE method fuses transferred priors and physics-based decomposition to enhance underwater images without paired labels.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.