What CVAT is
CVAT (Computer Vision Annotation Tool) is an open-source, MIT-licensed web application for labeling images and video. It was started at Intel and is now developed by CVAT.ai together with its community. Annotators draw bounding boxes, polygons, polylines, points, masks and skeletons, and track objects across video frames with interpolation, so a shape does not have to be redrawn on every frame.
CVAT can also call models while you label: interactive segmentation tools such as SAM turn a few clicks into a mask, which you then correct. Finished annotations export to common formats including COCO, YOLO and Pascal VOC.
How OpenSurgicalVision uses it
The project runs its own CVAT instance in a private Google Cloud environment in the EU. Annotation follows one shared class schema, and every label is reviewed before it can enter a dataset.
- Research use only. Labels serve research and are never clinical guidance.
- De-identified data only. Media reaches CVAT only after passing the de-identification pipeline; at this stage the workspace holds synthetic test data.
- Projects and tasks are prepared by administrators. Annotators open the jobs assigned to them, label and save; uploading new media from the browser is intentionally disabled.
Signing in
- Open the workspace with the button above.
- Sign in with the user name and password an administrator gave you. Only administrators create accounts; there is no self-registration.
- Keep your password private. Repeated failed sign-ins from one address are temporarily blocked.
Not invited yet? Medical students and educators: how to become an annotator. Everyone else: read how to contribute and the ethics framework, then ask the team through an issue. Never attach patient files to an issue or a message.