MambaIR is a simple yet effective baseline for image restoration using state-space models. The paper introduces MambaIR, which improves the standard Mamba model by incorporating local enhancement and channel attention to address issues like local pixel forgetting and channel redundancy. MambaIR consists of three stages: shallow feature extraction, deep feature extraction, and high-quality image reconstruction. The deep feature extraction stage uses Residual State-Space Blocks (RSSBs) to enhance local interactions and reduce channel redundancy. The model achieves a global effective receptive field with linear computational complexity, making it efficient for image restoration tasks. Extensive experiments show that MambaIR outperforms existing methods like SwinIR in image super-resolution, achieving up to 0.45dB improvement with similar computational cost. The model is also effective in image denoising, demonstrating its versatility in low-level vision tasks. The paper highlights the potential of state-space models, particularly Mamba, for efficient long-range modeling in image restoration, offering a new alternative to CNN and Transformer-based methods.MambaIR is a simple yet effective baseline for image restoration using state-space models. The paper introduces MambaIR, which improves the standard Mamba model by incorporating local enhancement and channel attention to address issues like local pixel forgetting and channel redundancy. MambaIR consists of three stages: shallow feature extraction, deep feature extraction, and high-quality image reconstruction. The deep feature extraction stage uses Residual State-Space Blocks (RSSBs) to enhance local interactions and reduce channel redundancy. The model achieves a global effective receptive field with linear computational complexity, making it efficient for image restoration tasks. Extensive experiments show that MambaIR outperforms existing methods like SwinIR in image super-resolution, achieving up to 0.45dB improvement with similar computational cost. The model is also effective in image denoising, demonstrating its versatility in low-level vision tasks. The paper highlights the potential of state-space models, particularly Mamba, for efficient long-range modeling in image restoration, offering a new alternative to CNN and Transformer-based methods.