Skip to content
Get packs

CVAT

Open-source, self-hosted tool for labeling images, video, audio and 3D data for AI training.

GitHub Download v2.81.0
Stars
17k
Forks
4k
Licence
MIT
Language
Python
Last update
4 days ago
Type
App

What it is

CVAT is a web-based tool for annotating datasets, meaning adding labels such as bounding boxes, polygons and masks to images, video, audio and 3D point clouds so machine learning models can learn from them. The Community edition is free and runs on your own servers via Docker, so your data stays in your infrastructure. Teams use it to split labeling work among annotators, review results, and export labeled data in formats like COCO or YOLO.

  • Draw bounding boxes or polygons on a set of images and export them in COCO or YOLO format
  • Upload video and track objects across frames using annotation tools
  • Assign labeling jobs to several annotators and track progress
  • Pre-label data with a connected serverless model such as SAM or YOLO v7, then correct it by hand
  • Script task creation and data uploads from Python with the cvat-sdk package

Great for

  • Computer vision teams that need to label images or video for model training
  • Organizations that must keep training data on their own infrastructure
  • Teams with several annotators who need task assignment and review
  • Engineers who want to automate labeling tasks through the Python SDK or REST API
  • Projects that plan to use their own models to pre-label data

Not ideal for

  • Someone who only needs to label a few images once: the Docker-based setup is more than a quick task requires, and a hosted option such as CVAT Online may be simpler
  • Users on Safari or other non-Chromium browsers: the README says CVAT is primarily tested with Chromium-based browsers and Safari/WebKit is not supported
  • Teams wanting a fully managed service without running infrastructure: the README points to CVAT Online or Enterprise for that, and some features are reserved for paid plans
Step-by-step guide: how to use CVATInstall, first run, examples, tips and FAQs

How to get it

  1. 1

    Clone and build from source

    You'll need: Docker Engine · Docker Compose · Git

    Clone the repository

    git clone https://github.com/cvat-ai/cvat

    Enter the project folder

    cd cvat

    Start the default stack

    docker compose up -d

    Create an admin account

    docker exec -it cvat_server bash -ic 'python3 ~/manage.py createsuperuser'
  2. 2

    Run it

    • Start the stack with docker compose up -d from the cvat folder.
    • Create an admin account with the createsuperuser command.
    • Open http://localhost:8080 in a Chromium-based browser and log in.
    • Create a project or task, upload your images or videos, and define labels.
  3. 3

    Get the most out of it

    • Set CVAT_HOST to your IP or domain if you want to reach the server from other machines
    • Connect serverless models by adding the serverless component to enable automatic annotation
    • Export labeled data in one of the 20+ supported formats such as COCO, YOLO or Pascal VOC
    • Use the Python SDK or CLI to script task creation, uploads and exports
    • Connect cloud storage such as S3, Azure or Google Cloud to load data from there

Downloads

Source code (.zip)develop branch · straight from GitHubRelease v2.81.0Release notes and sourceOfficial website & docswww.cvat.ai

Strengths

  • Supports many annotation types across image, video, audio and 3D data
  • Free, MIT-licensed Community edition that you can self-host and modify
  • Includes a Python SDK, CLI and REST API for automating workflows

Trade-offs

  • Self-hosting requires Docker and Docker Compose, which adds setup and maintenance work
  • Some advanced features, such as built-in SAM 2 and SAM 3 auto-labeling, AI agents and SSO, are listed as available only in paid CVAT Online or Enterprise plans
  • Browser support is limited to Chromium-based browsers as the primary tested option

How it compares

ProjectStarsTypeLevelUpdatedBest for
cvat (this)17kAppintermediate4 days agoComputer vision teams that need to label images or video for model training
labelImg25kDev toolbeginner2.4 years agoPeople who need a simple local tool to draw bounding boxes on a small image set
vit-pytorch25.6kLibraryintermediate5 days agoDevelopers who want to build an image classifier in PyTorch
supervision51.5kLibraryintermediate4 days agoPython developers building apps on top of object detection or segmentation models
carla14.6kAppadvanced4 days agoAutonomous driving researchers testing perception or control software
doccano10.8kAppbeginner7 months agoMachine learning teams building labeled text datasets

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.