Abstract:Although learning-based image restoration methods have made significant progress, they still struggle with limited generalization to real-world scenarios due to the substantial domain gap caused by training on synthetic data. Existing methods address this issue by improving data synthesis pipelines, estimating degradation kernels, employing deep internal learning, and performing domain adaptation and regularization. Previous domain adaptation methods have sought to bridge the domain gap by learning domain-invariant knowledge in either feature or pixel space. However, these techniques often struggle to extend to low-level vision tasks within a stable and compact framework. In this paper, we show that it is possible to perform domain adaptation via the noise space using diffusion models. In particular, by leveraging the unique property of how auxiliary conditional inputs influence the multi-step denoising process, we derive a meaningful diffusion loss that guides the restoration model in progressively aligning both restored synthetic and real-world outputs with a target clean distribution. We refer to this method as denoising as adaptation. To prevent shortcuts during joint training, we present crucial strategies such as channel-shuffling layer and residual-swapping contrastive learning in the diffusion model. They implicitly blur the boundaries between conditioned synthetic and real data and prevent the reliance of the model on easily distinguishable features. Experimental results on three classical image restoration tasks, namely denoising, deblurring, and deraining, demonstrate the effectiveness of the proposed method.

SLN-RED: Regularization by Simultaneous Local and Nonlocal Denoising for Image Restoration.

Frequency-Relevant Residual Learning for Multi-Modal Image Denoising.

The Little Engine that Could: Regularization by Denoising (RED)

Denoising Adversarial Networks for Rain Removal and Reflection Removal.

Monotonically Convergent Regularization by Denoising

Constrained Regularization by Denoising With Automatic Parameter Selection

Regularization by Denoising via Fixed-Point Projection (RED-PRO)

Dilated Residual Encode-Decode Networks for Image Denoising

Regularization by Denoising: Clarifications and New Interpretations

An Interpretation of Regularization by Denoising and its Application with the Back-Projected Fidelity Term

Deep Graph Laplacian Regularization for Robust Denoising of Real Images

Guided Regularization and its Application for Image Restoration

Image Denoising Via Bandwise Adaptive Modeling and Regularization Exploiting Nonlocal Similarity

On a nonlinear nonlocal reaction-diffusion system applied to image restoration

Online Regularization by Denoising with Applications to Phase Retrieval

Image Restoration with Mean-Reverting Stochastic Differential Equations

Averaged Deep Denoisers for Image Regularization

Denoising as Adaptation: Noise-Space Domain Adaptation for Image Restoration

Convolutional Neural Network Combined with Half-Quadratic Splitting Method for Image Restoration