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. Review collected by and hosted on G2.com.

The main thing we hit was that the default exports include only the frames that have been explicitly annotated. For single-frame detection that's fine, but it breaks sliding-window, multi-frame temporal models that expect dense, contiguous frames — we ended up building a dense frame-extraction step around the export to work around it. The keypoint workflow also has a slightly steeper learning curve than plain bounding-box work. None of these are dealbreakers, but they're worth planning for. Review collected by and hosted on G2.com.

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. Review collected by and hosted on G2.com.

No option for annotation around an object of no interest. For example, an object of no interest lies within an inner circle of an object of interest. I would like to exclude that object.

Shortcut for viewing and hiding the annotations during `Review` mode.

Slow loading times when opening projects, tasks, jobs.

No option for exporting issues together with annotations. Review collected by and hosted on G2.com.

What I particularly appreciate about CVAT is the ability to collaborate effectively with a team during the annotation phases; the platform really facilitates working together on the same AI project. The management of annotation classes is clear and well thought out, and the tool is generally smooth to use on a daily basis. Another strong point: the ability to import an already annotated dataset as a starting point, which allows you to gradually enrich your training data without starting from scratch. This is a considerable time saver in ML pipelines. Review collected by and hosted on G2.com.

One possible area for improvement concerns the task filtering system, which could be a bit more intuitive. When managing multiple projects in parallel, it takes some time to adapt to navigate effectively. It's not a blocker, but a simplified filtering interface would be an appreciated addition for teams working on large volumes. Review collected by and hosted on G2.com.

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. Review collected by and hosted on G2.com.

When we export large amount of data for the backup, the job may fail. Review collected by and hosted on G2.com.

I love using CVAT.ai because it gives us a seamless experience with our data annotation. Since our team is dispersed across different locations, it's much easier for everyone to log in to a central portal to do annotations. One of the key features that I like is the ability to pinpoint a specific object instance, create a hyperlink for it, and send it to my team with questions or queries regarding it, and vice versa. Review collected by and hosted on G2.com.

I think the user experience, especially the platform's loading speed, could be improved significantly. Sometimes it feels very slow. Also, I wish there was real-time analytics instead of having to refresh analytics all the time. On another note, during the initial setup, onboarding team members was tricky because they couldn't see any batches and tasks unless I made them admin. Review collected by and hosted on G2.com.

Several features in CVAT are very helpful for increasing productivity during labeling. First, the “Show bitmap” option is especially powerful for catching human errors, such as missing gaps between adjacent polygons. Another useful feature is the automatic bordering option, which helps when working with adjacent polygons by making it easier to add only the remaining polygon. Review collected by and hosted on G2.com.

There are a few features that could improve CVAT. First, being able to add new images to an existing task would be helpful, although it might introduce some drawbacks for the current data management workflow. Another useful improvement would be, during export, having the option to select multiple tasks from a project and/or exclude images that haven’t been labeled yet. This is still easy to fix in the exported file, but having this option built in could be very powerful and would make the process smoother. Review collected by and hosted on G2.com.

CVAT.ai has been very useful for our AI workflow at EVI Safety Technology, especially for video annotation in maritime safety scenarios.

The UI is intuitive, onboarding was quick, and the platform integrates well with our existing pipeline. Performance has been stable even with large video datasets, while AI-assisted annotation helps reduce manual work and speeds up dataset preparation significa Review collected by and hosted on G2.com.

Some advanced workflows and settings can still feel slightly technical for non-engineering users, especially during project configuration or task management at scale. Review collected by and hosted on G2.com.

I like pre-annotating data with CVAT.ai, it works really well. The platform is intuitive to use, and I quickly found what we intended to do. Compared to our self-built tool, CVAT.ai is simply incomparable. It greatly reduces the labeling effort, I estimate to a quarter to a fifth of what it would otherwise be. Overall, I find it really good, even if it might take a bit of time to get used to. Review collected by and hosted on G2.com.

Yes, so maybe it needs a bit of familiarization time, but overall I actually find it really good. Review collected by and hosted on G2.com.

I appreciate that their team is available and feels humble, genuinely looking to satisfy customers, and it's not complicated to get help. I also like their human approach. I love that they have student prices, making it accessible. The tool is user-friendly and easy to understand, even before reading their documentation, which makes it approachable. I also love the many options available to work with complementary tools like Segment Anything. Review collected by and hosted on G2.com.

Nothing! I disliked how zooming on images would sometimes block comments or options within, but after telling them about it, they resolved the issue. Review collected by and hosted on G2.com.

A good tool for MV annotation projects and development. Pricing is also affordable Review collected by and hosted on G2.com.

Maybe there should be more foundation models available, along with a “single-shot” annotation option, so only a few images are needed instead of thousands. Review collected by and hosted on G2.com.