MPRNet

Why we included this project

MPRNet is a CVPR 2021 architecture aimed at restoring degraded photographs, handling motion blur, rain streaks, and sensor noise in one framework. The network works in stages: it first pulls out global context with encoder-decoder blocks, then a high-resolution branch keeps the fine detail, with per-pixel attention reweighting features at each step. That staged design makes it a handy baseline if you are benchmarking your own restoration work, and the PyTorch repo includes training and testing scripts plus pretrained models for each task. It is research code rather than a plug-and-play library, so plan on reading and adapting the implementation yourself. For anyone reproducing CVPR results or learning how progressive restoration is put together, the stage organization and supervision scheme are worth studying.