openpose_train

Why we included this project

This is the training side of OpenPose, the real-time multi-person pose estimation system, and the place where the models that ship in the main project get trained. Researchers who want to reproduce or adapt the published results, rather than just load a pretrained model, will find the training scripts, validation tooling, and experimental model definitions here. The repo is explicitly experimental and not production ready, so it suits people comfortable building and tuning models themselves, not teams that need a drop-in inference service. It also carries experimental whole-body (BODY_135) and higher-accuracy BODY_25B models that never made it into the main OpenPose release, a useful reference for anyone studying single-network whole-body pose estimation.