Mamba-based All-in-One Image Restoration via Prompt
Huaizheng Lu, Dedong Zhang, Bin Huang · 2024
Image restoration is the process of restoring the original high-quality image from a degraded one. Currently, All-in-One image restoration methods have garnered significant attention due to their ability to remove different types of degradations within a single model. However, existing Transformer-based methods have issues such as high computational complexity and a lack of effective capture of long-range dependencies. To address this issue, this paper introduces the Mamba model, which designs a novel block to replace the existing Transformer-based blocks. Additionally, inspired by prompt learning, this paper also incorporates prompt learning to guide the image reconstruction. Through comparative experiments and ablation studies, we have demonstrated the feasibility and potential of the Mamba model in the field of All-in-One image restoration.