Learning to Restore Multi-Degraded Images via Ingredient Decoupling and Task-Aware Path Adaptation

TL;DR

This paper introduces IMDNet, an adaptive multi-degradation image restoration network designed to handle images with coexisting degradations by decoupling degradation ingredients and guiding restoration paths. The encoder uses the Degradation Ingredient Decoupling Block (DIDBlock) to separate degradation information across spatial and frequency domains, while the Fusion Block (FBlock) aggregates this information across levels, and the decoder employs the Task Adaptation Block (TABlock) to dynamically activate restoration branches based on the degradation representation. The approach is reinforced by a loss that enforces decoupling between clean and degradation features and balances spatial and frequency-domain cues. Across MDIR and SDIR benchmarks, IMDNet achieves state-of-the-art or competitive performance, with notable improvements in multi-degradation scenarios and strong generalization to individual degradation tasks, while maintaining computational efficiency.

Abstract

Image restoration (IR) aims to recover clean images from degraded observations. Despite remarkable progress, most existing methods focus on a single degradation type, whereas real-world images often suffer from multiple coexisting degradations, such as rain, noise, and haze coexisting in a single image, which limits their practical effectiveness. In this paper, we propose an adaptive multi-degradation image restoration network that reconstructs images by leveraging decoupled representations of degradation ingredients to guide path selection. Specifically, we design a degradation ingredient decoupling block (DIDBlock) in the encoder to separate degradation ingredients statistically by integrating spatial and frequency domain information, enhancing the recognition of multiple degradation types and making their feature representations independent. In addition, we present fusion block (FBlock) to integrate degradation information across all levels using learnable matrices. In the decoder, we further introduce a task adaptation block (TABlock) that dynamically activates or fuses functional branches based on the multi-degradation representation, flexibly selecting optimal restoration paths under diverse degradation conditions. The resulting tightly integrated architecture, termed IMDNet, is extensively validated through experiments, showing superior performance on multi-degradation restoration while maintaining strong competitiveness on single-degradation tasks.

Figures (7)

• Figure 1: The figure presents t-SNE visualizations of degradation embeddings from IMDNet (ours) and Defusion DefusionLuo_2025_CVPR. Our model exhibit clearer clustering, highlighting the effectiveness of decomposing degradation ingredients into decoupled representations.
• Figure 2: Mechanisms of our method. Our approach achieves multi-degradation image restoration by decomposing degradation ingredients into decoupled representations.
• Figure 3: (a) The overall architecture of the proposed IMDNet. (b) The structure of degradation ingredient decoupling block (DIDBlock). (c) The structure of task adaptation block (TABlock).
• Figure 4: Qualitative results under the MDIR experimental setup. Compared to other methods, our IMDNet effectively reduces color distortion and produces images that are visually closer to the ground truth.
• Figure 5: Qualitative results under the SDIR experimental setup. Our IMDNet recovers finer details in the reconstructed images.