Dynamic Degradation Decomposition Network for All-in-One Image Restoration

TL;DR

This paper tackles the challenge of restoring clean images from multiple, unknown degradations with a single model. It introduces D^3Net, which combines a restoration reconstruction branch with a degradation decomposition branch guided by prompts generated from a Cross-Domain Degradation Analyzer (CDDA) that fuses frequency-domain and spatial-domain information, and a Dynamic Decomposition Mechanism (DDM) that uses Gumbel-Softmax-driven Decision Units to adaptively activate Adaptive Decomposition Blocks. The key innovations are the two-level prompts ($C_t$ and $S_t$) derived from frequency-aware features and their use to drive a progressive, adaptive decomposition of degradation features, yielding superior PSNR/SSIM on Rain100L, SOTS-Outdoor, BSD68, GoPro, and LOL datasets, while also reducing computational overhead. The approach demonstrates strong generalization to unknown degradations and shows potential for scalable, real-world restoration tasks, with insights gathered from frequency-domain analyses and visualization studies like t-SNE. Overall, D^3Net advances all-in-one image restoration by leveraging cross-domain degradation modeling and prompt-guided dynamic processing, achieving state-of-the-art results and offering practical benefits for diverse imaging applications.

Abstract

Currently, restoring clean images from a variety of degradation types using a single model is still a challenging task. Existing all-in-one image restoration approaches struggle with addressing complex and ambiguously defined degradation types. In this paper, we introduce a dynamic degradation decomposition network for all-in-one image restoration, named D$^3$Net. D$^3$Net achieves degradation-adaptive image restoration with guided prompt through cross-domain interaction and dynamic degradation decomposition. Concretely, in D$^3$Net, the proposed Cross-Domain Degradation Analyzer (CDDA) engages in deep interaction between frequency domain degradation characteristics and spatial domain image features to identify and model variations of different degradation types on the image manifold, generating degradation correction prompt and strategy prompt, which guide the following decomposition process. Furthermore, the prompt-based Dynamic Decomposition Mechanism (DDM) for progressive degradation decomposition, that encourages the network to adaptively select restoration strategies utilizing the two-level prompt generated by CDDA. Thanks to the synergistic cooperation between CDDA and DDM, D$^3$Net achieves superior flexibility and scalability in handling unknown degradation, while effectively reducing unnecessary computational overhead. Extensive experiments on multiple image restoration tasks demonstrate that D$^3$Net significantly outperforms the state-of-the-art approaches, especially improving PSNR by 5.47dB and 3.30dB on the SOTS-Outdoor and GoPro datasets, respectively.

Figures (8)

• Figure 1: Conceptual differences between our work and other existing all-in-one methods (better viewed by zooming in).
• Figure 2: The architecture of D$^3$Net, which consists of a restoration reconstruction branch and a degradation decomposition branch.
• Figure 3: Dynamic decomposition mechanism: the upper part is adaptive decomposition blocks and the lower part is decision units.
• Figure 4: Qualitative comparison with the SOTA methods on the Rain100L, and GoPro datasets (better viewed by zooming in).
• Figure 5: Qualitative comparison with the SOTA methods on the BSD68, SOTS, and LOL datasets (better viewed by zooming in).