Self-supervised Dynamic Heterogeneous Degradation Modeling for Unified Zero-Shot Image Restoration

2026-05-23

AI Summary

Existing zero-shot image restoration methods often neglect physical consistency priors and rely on deep networks or pretrained features to model degradation, resulting in high training costs, fixed inference trajectories, and susceptibility to suboptimal solutions under complex degradations. This work proposes a Unified Physical Zero-shot Image Restoration framework (UP-ZeroIR), which, for the first time, explicitly reparameterizes heterogeneous degradations into a small set of physically consistent, compact parameters and models them in latent space as a unified all-in-one distribution. Additionally, it introduces dynamic diffusion trajectory adaptation and a self-supervised quality optimization mechanism, enabling adaptive restoration without task-specific training. The proposed method significantly outperforms existing zero-shot approaches under both single-type and mixed degradation scenarios, achieving state-of-the-art performance.

Abstract

Zero-shot image restoration provides a flexible way to handle diverse degradations without task-specific training. However, existing methods typically rely on stacked layers or pre-trained features to enhance degradation expression, while overlooking physically consistent priors. The insufficient degradation prompts impose the heavy training burden and high sampling costs during zero-shot diffusion. Moreover, the fixed inference trajectory often collapses to suboptimal solutions under complex corruptions. We observe that heterogeneous degradations can be reparameterized into a minimal set of physically coherent parameters for compact representation. Based on this insight, we first propose a unified physical zero-shot image restoration (UP-ZeroIR) framework that explicitly models heterogeneous degradations into a homogeneous all-in-one distribution. The distribution can be optimized directly in the latent space, enabling principled solution exploration and effective prompt adaptation. Besides, we introduce a dynamic quality-refinement strategy that adaptively adjusts the diffusion trajectory for robust globally optimal convergence. Extensive experiments demonstrate that our method achieves state-of-the-art performance across both single and mixed degradations. Our code is available at https://github.com/yangjinglyy/UP-ZeroIR

Problem

• zero-shot image restoration
• heterogeneous degradations
• diffusion trajectory
• degradation modeling
• physically consistent priors

Innovation

• zero-shot image restoration
• heterogeneous degradation modeling
• physical priors
• dynamic diffusion trajectory
• unified restoration framework