cvat
Annotate image, video, audio, and 3D datasets with manual tools and automated serverless AI models.
Label visual, audio, and 3D sensor data on your own infrastructure with team workflows and SDK access.
CVAT is a data labeling platform designed to prepare visual and spatial datasets for computer vision models. You can run it on your own servers to retain strict control over proprietary industrial training data. It suits engineering teams building computer vision systems for quality inspection, defect detection, or robotics.
What it is
The platform provides annotation tools across images, video streams, audio files, and 3D point clouds using bounding boxes, polygons, masks, keypoints, and cuboids. You can connect serverless machine learning models like SAM, YOLO v7, and RetinaNet via Nuclio for automated detection and tracking. It supports dataset management, multi-user assignment, quality control workflows, and exports to over 20 standard formats.
Who it's for
CVAT suits AI development teams, computer vision engineers, and dataset annotators who need full operational control over sensor data. It works well for industrial companies that require air-gapped deployments or cloud integrations with S3, Azure, or Google Cloud. It does not suit solo developers looking for zero-configuration desktop applications.
Key features
Multi-modal dataset support
Annotate images, video feeds, audio, and 3D point clouds using bounding boxes, polygons, masks, and cuboids.
Serverless auto-labeling
Connect ML models like SAM, YOLO v7, and TransT via Nuclio to automate detection, segmentation, and object tracking.
Broad format compatibility
Import and export dataset annotations across 20+ formats including COCO, YOLO, Pascal VOC, and KITTI.
Developer SDK and API
Automate task creation, file uploads, and dataset extractions using the Python SDK, CLI tool, or REST API.
Up and running in 3 steps
Our honest verdict
What's great
- Handles multi-modal data including 3D point clouds, video streams, and audio.
- Integrates serverless AI models via Nuclio to accelerate manual labeling tasks.
- Exports and imports datasets across 20+ computer vision formats out of the box.
- Includes role-based project workflows, task assignment, and issue tracking.
Where it falls short
- Requires managing Docker containers, databases, and server infrastructure.
- Enabling AI auto-labeling requires deploying and configuring a separate serverless Nuclio stack.
- Browser support is restricted to Chromium-based browsers like Chrome and Edge.
- Advanced UI analytics and SAM 2/3 features are restricted to paid hosted or enterprise plans.
Pricing & license
CVAT Community is free and open-source under the MIT license for self-hosted deployments. Commercial options include CVAT Online paid plans (Solo and Team) and CVAT Enterprise for organizations that want managed infrastructure or dedicated support.
Open remote AI-training roles
873 open remote AI-training roles, updated November 15, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
Use it, or have us set it up
FAQ
Which web browsers are supported?
CVAT is primarily tested on Chromium-based browsers like Google Chrome and Microsoft Edge. Firefox has limited support, while Safari and WebKit are unsupported.
How do I activate automated labeling models?
You must spin up the serverless component by running `docker compose -f docker-compose.yml -f components/serverless/docker-compose.serverless.yml up -d` to launch the Nuclio framework.
