Abstract
All-in-One image restoration aims to address multiple image degradationproblems using a single model, significantly reducing training costs anddeployment complexity compared to traditional methods that design dedicatedmodels for each degradation type. Existing approaches typically rely onDegradation-specific models or coarse-grained degradation prompts to guideimage restoration. However, they lack fine-grained modeling of degradationinformation and face limitations in balancing multi-task conflicts. To overcomethese limitations, we propose DPMambaIR, a novel All-in-One image restorationframework. By integrating a Degradation-Aware Prompt State Space Model (DP-SSM)and a High-Frequency Enhancement Block (HEB), DPMambaIR enables fine-grainedmodeling of complex degradation information and efficient global integration,while mitigating the loss of high-frequency details caused by task competition.Specifically, the DP-SSM utilizes a pre-trained degradation extractor tocapture fine-grained degradation features and dynamically incorporates theminto the state space modeling process, enhancing the model's adaptability todiverse degradation types. Concurrently, the HEB supplements high-frequencyinformation, effectively addressing the loss of critical details, such as edgesand textures, in multi-task image restoration scenarios. Extensive experimentson a mixed dataset containing seven degradation types show that DPMambaIRachieves the best performance, with 27.69dB and 0.893 in PSNR and SSIM,respectively. These results highlight the potential and superiority ofDPMambaIR as a unified solution for All-in-One image restoration.