From Controlled Scenarios to Real-World: Cross-Domain Degradation Pattern Matching for All-in-One Image Restoration

TL;DR

The paper tackles the domain gap that hinders All-in-One Image Restoration (AiOIR) in real-world conditions. It introduces Unified Domain-Adaptive Image Restoration (UDAIR), which combines a codebook of degradation prototypes, cross-sample contrastive learning (CSCL), and test-time adaptation (TTA) with a CORAL-based domain alignment to transfer knowledge from a source domain to target domains. Key contributions include the Degradation Aware and Analysis Module (DAAM) with a discretized codebook, a cross-domain latent space for degradation patterns, and an online adaptation mechanism that tightens target-domain features to source-domain clusters, achieving state-of-the-art results across five IR tasks on ten datasets. The approach demonstrates robust generalization to unseen degradations with improved restoration quality while maintaining efficiency, highlighting practical potential for real-world vision systems.

Abstract

As a fundamental imaging task, All-in-One Image Restoration (AiOIR) aims to achieve image restoration caused by multiple degradation patterns via a single model with unified parameters. Although existing AiOIR approaches obtain promising performance in closed and controlled scenarios, they still suffered from considerable performance reduction in real-world scenarios since the gap of data distributions between the training samples (source domain) and real-world test samples (target domain) can lead inferior degradation awareness ability. To address this issue, a Unified Domain-Adaptive Image Restoration (UDAIR) framework is proposed to effectively achieve AiOIR by leveraging the learned knowledge from source domain to target domain. To improve the degradation identification, a codebook is designed to learn a group of discrete embeddings to denote the degradation patterns, and the cross-sample contrastive learning mechanism is further proposed to capture shared features from different samples of certain degradation. To bridge the data gap, a domain adaptation strategy is proposed to build the feature projection between the source and target domains by dynamically aligning their codebook embeddings, and a correlation alignment-based test-time adaptation mechanism is designed to fine-tune the alignment discrepancies by tightening the degradation embeddings to the corresponding cluster center in the source domain. Experimental results on 10 open-source datasets demonstrate that UDAIR achieves new state-of-the-art performance for the AiOIR task. Most importantly, the feature cluster validate the degradation identification under unknown conditions, and qualitative comparisons showcase robust generalization to real-world scenarios.

Figures (12)

• Figure 1: To bridge the gap between closed scenarios and real-world scenarios, the proposed method establishes a cross-domain latent space via a codebook, which stores the shared intrinsic features of degradation patterns constrained through cross-sample contrastive learning. These features serve as prototypes of degradation patterns, effectively leveraging the knowledge and priors learned from closed scenarios. Test-time adaptation is further employed to fine-tune the model for each sample, compensating for errors introduced during cross-domain matching. This strategy mitigates performance decline caused by domain shifts in degradation pattern recognition and guides the network to restore high-quality images from low-quality inputs affected by various degradations.
• Figure 2: Overview of the proposed framework. The blue pipeline represent the dynamic domain adaptation strategy activated exclusively during target domain inference.
• Figure 3: Schematic diagram of Cross-Sample Contrastive Learning.
• Figure 4: Schematic diagram of Domain Adaptation Module.
• Figure 5: Overall comparisons on the source domain datasets.