# mit-han-lab/temporal-shift-module

[ICCV 2019] TSM: Temporal Shift Module for Efficient Video Understanding

Repository: https://github.com/mit-han-lab/temporal-shift-module
Canonical: https://ross.abutalabs.com/products/temporal-shift-module
Homepage: https://arxiv.org/abs/1811.08383
Language: Python
License: MIT
License Family: permissive
Topics: acceleration, low-latency, temporal-modeling, video-understanding, efficient-model, nvidia-jetson-nano, tsm
Last push: 2024-07-11T18:54:08+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2716, "days_push": 783, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2221, forks 421 (observed 2026-08-28T04:06:27.586763+00:00)

## What it is
PyTorch implementation of the Temporal Shift Module (TSM), an ICCV 2019 technique that adds temporal modeling to 2D CNNs at zero extra computation by shifting channels along the temporal dimension. It includes pretrained models on Kinetics and Something-Something and an online real-time demo for NVIDIA Jetson Nano.

## Use cases
- classify actions in videos efficiently
- run real-time video recognition on edge devices like Jetson Nano
- achieve 3D CNN accuracy with 2D CNN cost for video understanding
- fine-tune pretrained video models on custom datasets
- build online low-latency gesture or activity recognition
- perform online video object detection

## When to choose
- you need efficient temporal video modeling without 3D CNN compute
- deploying video recognition on low-power edge hardware
- you want strong pretrained checkpoints on Kinetics or Something-Something

## When to avoid
- you need actively maintained code with recent framework support
- your task is image-only with no temporal component
- you need a modern video model ecosystem beyond this paper's scope

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, computer-vision, video-processing, deep-learning
- domain: computer-vision, machine-learning, deep-learning
- platform: python
- tags: temporal-shift-module, video-understanding, pytorch, efficient-inference, edge-deployment, jetson-nano, kinetics, iccv-2019, pretrained-models, real-time, video, gpu, linux

## Member repositories
- mit-han-lab/temporal-shift-module (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:27.586763+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 2026-08-30T02:45:47.619627+00:00, confidence not recorded.
  - readme: https://github.com/mit-han-lab/temporal-shift-module (fetched 2026-08-28T04:06:27.586763+00:00, sha cec6c87e6dca)
  - homepage: https://arxiv.org/abs/1811.08383 (fetched 2026-08-29T10:26:05.576802+00:00, sha 1f36c97ccc4d)
  - site_page: https://info.arxiv.org/about/donate.html (fetched 2026-08-29T10:26:05.579486+00:00, sha cca9c3a11c56)
  - site_page: https://info.arxiv.org/about/ourmembers.html (fetched 2026-08-29T10:26:05.582785+00:00, sha 47cbc55ff1de)
  - site_page: https://info.arxiv.org/about (fetched 2026-08-29T10:26:05.584622+00:00, sha a1f16f915a9a)
  - site_page: https://info.arxiv.org/labs/index.html (fetched 2026-08-29T10:26:05.581124+00:00, sha b14a8d05a0ec)
- Data as of 2026-08-30T08:39:29.467469+00:00.