Efficient Degradation-aware Any Image Restoration

TL;DR

DaAIR tackles the problem of robust All-in-One image restoration by introducing a degradation-aware learner (DaLe) that jointly models shared and degradation-specific features in a low-rank space. A degradation-aware routing mechanism assigns capacity to specialized experts per degradation, while a self-learnable controller in the decoder uses accumulated encoder information to refine reconstruction with minimal overhead. This combination yields state-of-the-art performance across three and five degradations, outperforming task-specific and prior All-in-One models while significantly reducing parameters and GMACs. The approach offers practical benefits for real-world deployment by delivering high fidelity restorations at reduced computational cost, with strong potential for extensions using external inductive biases and more realistic degradation settings.

Abstract

Reconstructing missing details from degraded low-quality inputs poses a significant challenge. Recent progress in image restoration has demonstrated the efficacy of learning large models capable of addressing various degradations simultaneously. Nonetheless, these approaches introduce considerable computational overhead and complex learning paradigms, limiting their practical utility. In response, we propose \textit{DaAIR}, an efficient All-in-One image restorer employing a Degradation-aware Learner (DaLe) in the low-rank regime to collaboratively mine shared aspects and subtle nuances across diverse degradations, generating a degradation-aware embedding. By dynamically allocating model capacity to input degradations, we realize an efficient restorer integrating holistic and specific learning within a unified model. Furthermore, DaAIR introduces a cost-efficient parameter update mechanism that enhances degradation awareness while maintaining computational efficiency. Extensive comparisons across five image degradations demonstrate that our DaAIR outperforms both state-of-the-art All-in-One models and degradation-specific counterparts, affirming our efficacy and practicality. The source will be publicly made available at https://eduardzamfir.github.io/daair/

Figures (10)

• Figure 1: (a) Prior work employ inefficient models and learning for capturing degradation dependencies. (b) We route model capacity to efficiently learn shared and specialized embeddings.
• Figure 2: Model complexity. Our proposed DaAIR surpasses prior methods, achieving state-of-the-art results in All-in-One image restoration with enhanced efficiency.
• Figure 3: Architecture overview. DaAIR reconstructs missing information using an asymmetric encoder-decoder architecture. Each encoder block integrates our proposed Degradation-aware Learner (DaLe) adaptively routing model capacity. Additionally, the decoder blocks are complemented by a controller (C) to enhance overall reconstruction.
• Figure 4: Visual comparison of DaAIR with state-of-the-art methods on challenging cases for the All-in-One setting considering three degradations.
• Figure 4: Impact of key components. PSNR (dB, $\uparrow$) and SSIM ($\uparrow$) metrics are reported on the full RGB images.