CVAT

CVAT (Computer Vision Annotation Tool) is one of the best-known open-source data annotation platforms, originally open-sourced by Intel and now maintained by the CVAT.ai team. The free self-hosted Community edition covers image, video, and 3D annotation with team collaboration, task and review workflows, cloud storage integration, and developer SDKs and APIs, and the MIT-licensed core keeps your data entirely inside your own infrastructure.

It supports AI-assisted labeling by connecting your own models for detection, segmentation, and tracking, exports to standard formats like COCO, YOLO, and Pascal VOC, and underpins the commercial CVAT Online and Enterprise products. The project reports millions of Docker pulls and broad adoption across research and production vision teams.

Honest limitations: you are the ops team; uptime, scaling, QA policy, and compliance (SOC 2, HIPAA) are all yours to build. Reviewers note it can feel single-user past roughly ten thousand frames, with review queues and collaboration needing workarounds, and there is no native model registry or experiment tracking, so it pairs best with an external MLOps stack.

Deploy via Docker Compose from the GitHub repo, or sign up for the managed CVAT Online service.