Video

An open-source toolbox for video understanding based on PyTorch

MMAction2

MMAction2 is an open-source toolbox for video understanding based on PyTorch. It is a part of the OpenMMLab project.

The master branch works with PyTorch 1.3+.
mmaction2_overview
Action Recognition Results on Kinetics-400

spatio-temporal-det
Spatio-Temporal Action Detection Results on AVA-2.1

123989146-2ecae680-d9fb-11eb-916b-b9db5563a9e5
Skeleton-base Action Recognition Results on NTU-RGB+D-120

Major Features

  • Modular design

    We decompose the video understanding framework into different components and one can easily construct a customized video understanding framework by combining different modules.

  • Support for various datasets

    The toolbox directly supports multiple datasets, UCF101, Kinetics-[400/600/700], Something-Something V1&V2, Moments in Time, Multi-Moments in Time, THUMOS14, etc.

  • Support for multiple video understanding frameworks

    MMAction2 implements popular frameworks for video understanding: