BasicSR

Why we included this project

BasicSR bundles the standard image and video restoration models into one PyTorch codebase, so EDSR, RCAN, ESRGAN, EDVR, BasicVSR, SwinIR and similar architectures all share the same data loading and training infrastructure instead of each needing its own reimplementation. It handles super-resolution, denoising, deblurring, and JPEG artifact removal, and ships with ready-made training and inference commands, dataset preparation scripts, and a model zoo of pretrained weights. The toolbox is also importable as a Python package, not just a command-line tool, which makes it easy to fold into your own experiments. For anyone doing low-level vision research who wants a maintained reference implementation covering both single-image and video methods, this is a practical place to start.