# facebookresearch/TimeSformer

The official pytorch implementation of our paper "Is Space-Time Attention All You Need for Video Understanding?"

Repository: https://github.com/facebookresearch/TimeSformer
Canonical: https://ross.abutalabs.com/products/timesformer
Language: Python
License: NOASSERTION
License Family: other
Archived: true
Last push: 2024-04-09T06:00:55+00:00

## Health v2 (maintenance only)
Score: 10/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 1979, "days_push": 876, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, archived, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1863, forks 246 (observed 2026-08-28T04:05:46.035075+00:00)

## What it is
Official PyTorch implementation of TimeSformer, a space-time attention transformer model for video understanding from an ICML 2021 paper. It provides training and testing code plus pretrained models for video action recognition benchmarks like Kinetics-400.

## Use cases
- classify human actions in videos
- train a video transformer on Kinetics-400
- fine-tune a pretrained video model on my own action recognition dataset
- run inference with a pretrained video classification model
- compare video transformer architectures for research
- evaluate video models on Something-Something-V2

## When to choose
- you need a transformer-based video action recognition model with pretrained checkpoints
- you want a research baseline for video understanding experiments
- you need PyTorch code for training video classifiers on standard benchmarks

## When to avoid
- you need lightweight or real-time video classification on edge devices
- you want a production-ready video analytics service rather than research code
- you need video tasks beyond classification such as detection or tracking

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, video-processing
- domain: machine-learning, computer-vision
- platform: python
- tags: video-understanding, transformer, action-recognition, pytorch, research-code, icml-2021, video, gpu

## Member repositories
- facebookresearch/TimeSformer (main) score 10

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:46.035075+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 2026-08-30T03:15:38.801247+00:00, confidence not recorded.
  - readme: https://github.com/facebookresearch/TimeSformer (fetched 2026-08-28T04:05:46.035075+00:00, sha 868610d11980)
- Data as of 2026-08-30T08:39:29.467469+00:00.