mmskeleton

Why we included this project

If your work involves reading human movement from body joints rather than pixels, this toolbox gives you a solid base. It ships a complete pipeline for skeleton-based action recognition built around the ST-GCN graph convolutional network, and it also handles 2D pose estimation. The framework is built to be extended, so you can add your own datasets or build a custom application on top, and being part of the OpenMMLab family means the code follows a modular design that researchers and engineers in that ecosystem will recognize. Teams doing motion analysis for fitness, sports, security, or human-robot interaction can use it as a concrete starting point. Development has settled, so treat it as a stable reference implementation and baseline rather than a fast-moving framework.