The paper introduces a novel multi-stage progressive image restoration architecture called MPRNet, which aims to balance spatial details and high-level contextual information in recovering degraded images. The key contributions include: 1. **Multi-Stage Architecture**: MPRNet breaks down the image restoration task into multiple stages, each learning restoration functions for degraded inputs. This approach allows for more manageable and effective recovery processes. 2. **Encoder-Decoder and High-Resolution Branch**: The model first learns contextualized features using encoder-decoder architectures and then combines them with a high-resolution branch to retain local information. 3. **Supervised Attention Module (SAM)**: A per-pixel adaptive design that leverages in-situ supervised attention to reweight local features at each stage. 4. **Cross-Stage Feature Fusion (CSFF)**: A mechanism to propagate intermediate features from earlier stages to later stages, ensuring information exchange and stability in network optimization. 5. **Performance on Diverse Datasets**: MPRNet achieves state-of-the-art performance on ten datasets across various tasks including image deraining, deblurring, and denoising. The paper also discusses the limitations of existing single-stage approaches and the benefits of multi-stage designs, providing detailed ablations and qualitative results to support the effectiveness of MPRNet.The paper introduces a novel multi-stage progressive image restoration architecture called MPRNet, which aims to balance spatial details and high-level contextual information in recovering degraded images. The key contributions include: 1. **Multi-Stage Architecture**: MPRNet breaks down the image restoration task into multiple stages, each learning restoration functions for degraded inputs. This approach allows for more manageable and effective recovery processes. 2. **Encoder-Decoder and High-Resolution Branch**: The model first learns contextualized features using encoder-decoder architectures and then combines them with a high-resolution branch to retain local information. 3. **Supervised Attention Module (SAM)**: A per-pixel adaptive design that leverages in-situ supervised attention to reweight local features at each stage. 4. **Cross-Stage Feature Fusion (CSFF)**: A mechanism to propagate intermediate features from earlier stages to later stages, ensuring information exchange and stability in network optimization. 5. **Performance on Diverse Datasets**: MPRNet achieves state-of-the-art performance on ten datasets across various tasks including image deraining, deblurring, and denoising. The paper also discusses the limitations of existing single-stage approaches and the benefits of multi-stage designs, providing detailed ablations and qualitative results to support the effectiveness of MPRNet.