# open-mmlab/mmaction2

OpenMMLab's Next Generation Video Understanding Toolbox and Benchmark

Repository: https://github.com/open-mmlab/mmaction2
Canonical: https://ross.abutalabs.com/products/mmaction2
Homepage: https://mmaction2.readthedocs.io
Language: Python
License: Apache-2.0
License Family: permissive
Topics: action-recognition, temporal-action-localization, pytorch, video-understanding, tsn, i3d, slowfast, ava, spatial-temporal-action-detection, benchmark, tsm, x3d, non-local, deep-learning, openmmlab, posec3d, video-classification, uniformerv2
Last push: 2026-03-18T04:04:49+00:00

## Health v2 (maintenance only)
Score: 55/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 72, release rhythm 8, longevity 100
- inputs: {"age_days": 2244, "days_push": 168, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 5142, forks 1357 (observed 2026-08-28T04:09:11.148157+00:00)

## What it is
MMAction2 is OpenMMLab's PyTorch-based toolbox and benchmark for video understanding, covering action recognition, temporal action localization, and spatio-temporal action detection. It provides modular components, a large model zoo (TSN, I3D, SlowFast, TSM, X3D, NonLocal, PoseC3D, UniFormerV2), and training/evaluation pipelines.

## Use cases
- classify human actions in videos
- train a SlowFast or I3D model on Kinetics
- detect when actions occur in untrimmed videos
- run spatio-temporal action detection on AVA
- benchmark video classification models
- fine-tune a pretrained video model on my own dataset
- skeleton-based action recognition with PoseC3D

## When to choose
- you need state-of-the-art video action recognition or localization models in PyTorch
- you want a modular, configurable training framework with a large pretrained model zoo
- you are benchmarking video understanding approaches on standard datasets

## When to avoid
- you only need simple video encoding/decoding or editing rather than ML
- you work outside PyTorch or need lightweight inference on edge devices without GPU support
- you need general image-only vision tasks better served by mmdetection or timm

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, computer-vision, video-processing, benchmarking
- domain: computer-vision, deep-learning, machine-learning
- platform: python
- tags: action-recognition, video-understanding, pytorch, openmmlab, model-zoo, temporal-action-localization, spatio-temporal-action-detection, video, linux, gpu

## Member repositories
- open-mmlab/mmaction2 (main) score 55

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:11.148157+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:01:59.666803+00:00, confidence not recorded.
  - readme: https://github.com/open-mmlab/mmaction2 (fetched 2026-08-28T04:09:11.148157+00:00, sha 9defae280cab)
  - registry_pypi: https://pypi.org/pypi/mmaction2/json (fetched 2026-08-29T08:56:52.981771+00:00, sha 5865ffe41e5e)
- Data as of 2026-08-30T08:39:29.467469+00:00.