/mmaction2

OpenMMLab's Next Generation Video Understanding Toolbox and Benchmark

Primary LanguagePythonApache License 2.0Apache-2.0

Introduction

English | 简体中文

Documentation actions codecov PyPI LICENSE Average time to resolve an issue Percentage of issues still open

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+.


Action Recognition Results on Kinetics-400


Skeleton-base Action Recognition Results on NTU-RGB+D-120


Spatio-Temporal Action Detection Results on AVA-2.1

Major Features

  • Modular design: We decompose a video understanding framework into different components. One can easily construct a customized video understanding framework by combining different modules.

  • Support four major video understanding tasks: MMAction2 implements various algorithms for multiple video understanding tasks, including action recognition, action localization, spatio-temporal action detection, and skeleton-based action detection. We support 27 different algorithms and 20 different datasets for the four major tasks.

  • Well tested and documented: We provide detailed documentation and API reference, as well as unit tests.

Updates

  • (2021-11-24) We support 2s-AGCN on NTU60 XSub, achieve 86.06% Top-1 accuracy on joint stream and 86.89% Top-1 accuracy on bone stream respectively.
  • (2021-10-29) We provide a demo for skeleton-based and rgb-based spatio-temporal detection and action recognition (demo/demo_video_structuralize.py).
  • (2021-10-26) We train and test ST-GCN on NTU60 with 3D keypoint annotations, achieve 84.61% Top-1 accuracy (higher than 81.5% in the paper).
  • (2021-10-25) We provide a script(tools/data/skeleton/gen_ntu_rgbd_raw.py) to convert the NTU60 and NTU120 3D raw skeleton data to our format.
  • (2021-10-25) We provide a guide on how to train PoseC3D with custom datasets, bit-scientist authored this PR!
  • (2021-10-16) We support PoseC3D on UCF101 and HMDB51, achieves 87.0% and 69.3% Top-1 accuracy with 2D skeletons only. Pre-extracted 2D skeletons are also available.

Release: v0.20.0 was released in 30/10/2021. Please refer to changelog.md for details and release history.

Installation

Please refer to install.md for installation.

Get Started

Please see getting_started.md for the basic usage of MMAction2. There are also tutorials:

A Colab tutorial is also provided. You may preview the notebook here or directly run on Colab.

Supported Methods

Action Recognition
C3D (CVPR'2014) TSN (ECCV'2016) I3D (CVPR'2017) I3D Non-Local (CVPR'2018) R(2+1)D (CVPR'2018)
TRN (ECCV'2018) TSM (ICCV'2019) TSM Non-Local (ICCV'2019) SlowOnly (ICCV'2019) SlowFast (ICCV'2019)
CSN (ICCV'2019) TIN (AAAI'2020) TPN (CVPR'2020) X3D (CVPR'2020) OmniSource (ECCV'2020)
MultiModality: Audio (ArXiv'2020) TANet (ArXiv'2020) TimeSformer (ICML'2021)
Action Localization
SSN (ICCV'2017) BSN (ECCV'2018) BMN (ICCV'2019)
Spatio-Temporal Action Detection
ACRN (ECCV'2018) SlowOnly+Fast R-CNN (ICCV'2019) SlowFast+Fast R-CNN (ICCV'2019) LFB (CVPR'2019)
Skeleton-based Action Recognition
ST-GCN (AAAI'2018) PoseC3D (ArXiv'2021)

Results and models are available in the README.md of each method's config directory. A summary can be found on the model zoo page.

We will keep up with the latest progress of the community and support more popular algorithms and frameworks. If you have any feature requests, please feel free to leave a comment in Issues.

Supported Datasets

Action Recognition
HMDB51 (Homepage) (ICCV'2011) UCF101 (Homepage) (CRCV-IR-12-01) ActivityNet (Homepage) (CVPR'2015) Kinetics-[400/600/700] (Homepage) (CVPR'2017)
SthV1 (Homepage) (ICCV'2017) SthV2 (Homepage) (ICCV'2017) Diving48 (Homepage) (ECCV'2018) Jester (Homepage) (ICCV'2019)
Moments in Time (Homepage) (TPAMI'2019) Multi-Moments in Time (Homepage) (ArXiv'2019) HVU (Homepage) (ECCV'2020) OmniSource (Homepage) (ECCV'2020)
FineGYM (Homepage) (CVPR'2020)
Action Localization
THUMOS14 (Homepage) (THUMOS Challenge 2014) ActivityNet (Homepage) (CVPR'2015)
Spatio-Temporal Action Detection
UCF101-24* (Homepage) (CRCV-IR-12-01) JHMDB* (Homepage) (ICCV'2015) AVA (Homepage) (CVPR'2018)
Skeleton-based Action Recognition
PoseC3D-FineGYM (Homepage) (ArXiv'2021) PoseC3D-NTURGB+D (Homepage) (ArXiv'2021) PoseC3D-UCF101 (Homepage) (ArXiv'2021) PoseC3D-HMDB51 (Homepage) (ArXiv'2021)

Datasets marked with * are not fully supported yet, but related dataset preparation steps are provided. A summary can be found on the Supported Datasets page.

Benchmark

To demonstrate the efficacy and efficiency of our framework, we compare MMAction2 with some other popular frameworks and official releases in terms of speed. Details can be found in benchmark.

Data Preparation

Please refer to data_preparation.md for a general knowledge of data preparation. The supported datasets are listed in supported_datasets.md

FAQ

Please refer to FAQ for frequently asked questions.

Projects built on MMAction2

Currently, there are many research works and projects built on MMAction2 by users from community, such as:

  • Video Swin Transformer. [paper][github]
  • Evidential Deep Learning for Open Set Action Recognition, ICCV 2021 Oral. [paper][github]
  • Rethinking Self-supervised Correspondence Learning: A Video Frame-level Similarity Perspective, ICCV 2021 Oral. [paper][github]

etc., check projects.md to see all related projects.

License

This project is released under the Apache 2.0 license.

Citation

If you find this project useful in your research, please consider cite:

@misc{2020mmaction2,
    title={OpenMMLab's Next Generation Video Understanding Toolbox and Benchmark},
    author={MMAction2 Contributors},
    howpublished = {\url{https://github.com/open-mmlab/mmaction2}},
    year={2020}
}

Contributing

We appreciate all contributions to improve MMAction2. Please refer to CONTRIBUTING.md in MMCV for more details about the contributing guideline.

Acknowledgement

MMAction2 is an open-source project that is contributed by researchers and engineers from various colleges and companies. We appreciate all the contributors who implement their methods or add new features and users who give valuable feedback. We wish that the toolbox and benchmark could serve the growing research community by providing a flexible toolkit to reimplement existing methods and develop their new models.

Projects in OpenMMLab

  • MMCV: OpenMMLab foundational library for computer vision.
  • MIM: MIM Installs OpenMMLab Packages.
  • MMAction2: OpenMMLab's next-generation video understanding toolbox and benchmark.
  • MMClassification: OpenMMLab image classification toolbox and benchmark.
  • MMDetection: OpenMMLab detection toolbox and benchmark.
  • MMDetection3D: OpenMMLab's next-generation platform for general 3D object detection.
  • MMEditing: OpenMMLab image and video editing toolbox.
  • MMFewShot: OpenMMLab few shot learning toolbox.
  • MMFlow: OpenMMLab Optical Flow Toolbox and Benchmark.
  • MMGeneration: OpenMMLab image and video generative models toolbox.
  • MMHuman3D: OpenMMLab human pose and shape estimation toolbox and benchmark.
  • MMOCR: A Comprehensive Toolbox for Text Detection, Recognition and Understanding.
  • MMPose: OpenMMLab pose estimation toolbox and benchmark.
  • MMSegmentation: OpenMMLab semantic segmentation toolbox and benchmark.
  • MMTracking: OpenMMLab video perception toolbox and benchmark.