cvpods

Why we included this project

cvpods is a PyTorch codebase from Megvii that covers most of what a vision research group works on: classification, detection, segmentation, keypoints, self-supervised learning, and 3D tasks. The appeal is that you keep one experiment-management layer and swap between tasks instead of maintaining a separate training pipeline for each problem, which matters when a team runs several research directions at once. It ships a model zoo and a step-by-step tutorial, so reproducing published results or adapting a config to your own data is mostly a matter of editing a config file rather than rewriting code. It is a research toolbox.