CVAT

CVAT is a data annotation platform for teams preparing datasets for computer vision and audio models. It combines labeling tools with shared assignments and review workflows.

• Manual annotation: Label objects with boxes, polygons, masks, keypoints and 3D cuboids, track them through video, or annotate audio with intervals and tags.
• AI-assisted labeling: Use custom models or Community integrations such as Segment Anything, YOLO, RetinaNet and HRNet for detection, segmentation, pose estimation and tracking.
• Team review: Divide datasets into projects, tasks and jobs, assign annotators and reviewers, and discuss annotation issues with role-based access.
• Quality checks: Compare annotators' results and assess labels with ground truth and honeypot checks; Community exposes these checks through its server API.
• Dataset exchange: Import and export COCO, YOLO, Pascal VOC and KITTI datasets, connect local or cloud storage, and automate work through the API, Python SDK or CLI.

The open-source Community edition runs on your infrastructure with Docker under the MIT license. CVAT Online is hosted, while Enterprise runs in an organization's own cloud or internal environment. Community supports custom automatic annotation through Nuclio, with PyTorch, ONNX, OpenVINO and TensorFlow models. The browser interface is primarily tested with Chrome and Edge; Firefox has caveats, and Safari/WebKit isn't supported.