CVAT Open source

CVAT (Computer Vision Annotation Tool) is a web-based platform for labeling images, video, and 3D point-cloud data used to train computer vision models. It is developed by CVAT.ai and released under the permissive MIT license, so the core application can be self-hosted, modified, and redistributed at no cost.

What it does

The tool covers the common supervised-learning tasks: object detection, image classification, and semantic or instance segmentation. Annotators can draw bounding boxes, polygons, polylines, points, and keypoints, work with 3D cuboids for point clouds, and track objects across video frames using interpolation. AI-assisted labeling lets teams connect their own detection, segmentation, or tracking models to pre-label data and increase throughput.

Key features

• Image, video, and 3D point-cloud annotation in one interface
• Model-assisted labeling, automatic annotation, and object tracking
• Quality assurance, review, and team collaboration workflows
• Import and export across 20+ formats including COCO, YOLO, Pascal VOC, and KITTI
• REST API and SDK for programmatic access and pipeline integration

Deployment and licensing

The self-hosted community edition is free and open source under MIT and can run on a workstation or on-premises infrastructure via Docker. CVAT.ai also operates a hosted SaaS version and enterprise tiers that add support and services. The tooling is widely used by computer-vision engineers, research groups, and annotation teams that want full control over their data or need to avoid per-seat cloud costs.