openpifpaf

Why we included this project

OpenPifPaf is a good fit when your vision work involves locating people, animals, or objects and their key points in images or video. It's a PyTorch implementation of the Composite Fields method, so it detects body joints, faces, hands, and feet in a single pass and associates them across frames for tracking. That single-stage design keeps it fast enough for near real time use. Pretrained checkpoints also cover car and animal models, and the command line interface makes quick predictions easy. The codebase doubles as a readable reference for the CVPR 2019 and 2021 papers if you want to see how the method works.