- Bounding box or keypoint selection works well and intuitively, even when boxes and keypoints are clustered in a complex manner.<br>- Image loading is fast (1280 × 720 pixels).<br>- Object tracking interpolation across frames saves a lot of time.

Ease of use, simplicity, and cost for the value obtained

I really like the AI models in CVAT because they make meticulous annotation work easy with the latest models available. They are fairly accurate with our datasets, and even when they're not, we can easily tweak them. The export feature is crucial for us as it allows us to use annotations collected on the software and pick the format of the export, which is useful. CVAT allows us to share our datasets among annotators worldwide, link our data in AWS securely and quickly, review annotations with ease, and efficiently bring on and off annotators. I use CVAT with AWS S3 storage to load the images, and I find the integration quite smooth.