mmaction2 and mmskeleton
These tools are **ecosystem siblings** within the OpenMMLab framework; MMSkeleton is a toolbox specifically for skeleton-based methods, including pose estimation, while MMAction2 is a broader, next-generation video understanding toolbox that likely incorporates or supersedes many of the functionalities and research directions of MMSkeleton for action recognition, with MMAction2 being the more actively developed and widely adopted solution.
About mmaction2
open-mmlab/mmaction2
OpenMMLab's Next Generation Video Understanding Toolbox and Benchmark
Supports five major video understanding tasks—action recognition, localization, spatio-temporal detection, skeleton-based detection, and video retrieval—with modular architecture built on PyTorch, MMCV, and MMEngine. Implements both 2D/3D CNN architectures (TSN, TSM, SlowFast) and multi-modal approaches (ActionClip, VindLU) for temporal reasoning in videos. Integrates with MMDetection and MMPose for object detection and pose estimation pipelines.
About mmskeleton
open-mmlab/mmskeleton
A OpenMMLAB toolbox for human pose estimation, skeleton-based action recognition, and action synthesis.
Scores updated daily from GitHub, PyPI, and npm data. How scores work