# PaddlePaddle/PaddleVideo

Awesome video understanding toolkits based on PaddlePaddle. It supports video data annotation tools, lightweight RGB and skeleton based action recognition model, practical applications for video tagging and sport action detection.

Repository: https://github.com/PaddlePaddle/PaddleVideo
Canonical: https://ross.abutalabs.com/products/paddlevideo
Language: Python
License: Apache-2.0
License Family: permissive
Topics: video-recognition, tsm, slowfast, tsn, bmn, action-recognition, youtube-8m, kinetics400, video-understanding, activitynet, action-detection, temporal-action-detection, action-localization, ava, actbert, pp-tsm, videotag, st-gcn, t2vlad
Last push: 2025-02-12T06:53:52+00:00

## Health v2 (maintenance only)
Score: 26/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 6, release rhythm 8, longevity 100
- inputs: {"age_days": 2120, "days_push": 567, "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 1702, forks 387 (observed 2026-08-28T04:05:24.412984+00:00)

## What it is
PaddleVideo is a video understanding toolkit built on PaddlePaddle, offering state-of-the-art models for action recognition, temporal action detection, and video tagging. It covers the full pipeline from data annotation and training to model compression and deployment, including lightweight models like PP-TSM and skeleton-based models like ST-GCN.

## Use cases
- recognize human actions in videos
- train an action recognition model on Kinetics-400
- detect temporal action boundaries in sports footage
- classify videos with skeleton-based pose models
- deploy a lightweight video model on CPU
- annotate video datasets for training
- fine-tune video models for industrial or medical scenarios

## When to choose
- you use the PaddlePaddle ecosystem and need video understanding models
- you need a full pipeline from annotation to deployment for video tasks
- you want lightweight CPU-friendly action recognition models
- you work with skeleton-based action recognition or sports analytics

## When to avoid
- your stack is PyTorch or TensorFlow based
- you need general video editing or transcoding rather than understanding
- you need real-time streaming video analytics out of the box

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, video-processing, computer-vision, llm-training
- domain: deep-learning, computer-vision, machine-learning
- platform: python
- tags: video-understanding, action-recognition, action-detection, paddlepaddle, model-zoo, tsm, slowfast, st-gcn, video-tagging, skeleton-based, video, linux, gpu

## Member repositories
- PaddlePaddle/PaddleVideo (main) score 26

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:24.412984+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 2026-08-30T03:37:39.520928+00:00, confidence not recorded.
  - readme: https://github.com/PaddlePaddle/PaddleVideo (fetched 2026-08-28T04:05:24.412984+00:00, sha 313c5eb78edd)
  - registry_pypi: https://pypi.org/pypi/paddlevideo/json (fetched 2026-08-29T11:12:00.691296+00:00, sha 8dd428b3fc91)
- Data as of 2026-08-30T08:39:29.467469+00:00.