The biggest win for us is the model-assisted annotation loop. We run YOLO and SAM2 pre-annotation directly through CVAT, so the work is mostly verifying and correcting rather than labeling from scratch. For a small operation producing the computer-vision training data behind our product, that's the difference between a dataset taking weeks versus days. We also rely on the flexibility of handling multiple annotation types in one tool — we annotate ball and equipment bounding boxes, hockey stick parts, and pose keypoints, and the project/task structure keeps dozens of video tasks organized without friction. Being cloud-hosted means there's no annotation infrastructure for us to stand up or maintain, and the export formats drop straight into our YOLO/training pipeline.

I find CVAT.ai convenient to use between our core team and our outsourced annotators. It supports specific shapes that other similar tools do not support. The export and API access is reliable. While currently we enjoy the SaaS platform, it's good to know that they are open-source, and when there's financial justification, we can set up our own instance with minimum transition cost. I particularly appreciate the ease of use and the open-source nature. The annotation process is quick and reliable, and the workflow is managed. Setting up CVAT.ai was very easy.

I use this powerful annotation tool for SEM microscopy image annotation. I like the extensive support of prelabelling through foundation models or custom models via API. I use quality tools for reviewing, like quality reports and create issues with a review mode. The ui is nice and there are helpful shortcuts. It is worth the money. Performance is fine, but could be improved and there is the cvat documentation available online, it could be slightly improved as well.