CVAT tutorial: image labeling and dataset export
Learn to create a CVAT task, draw bounding boxes, assign classes and export a dataset as a ZIP with optional images.
Player not loading? Watch on YouTube
This beginner CVAT tutorial follows a browser workflow for preparing annotated computer vision datasets. The speaker starts with account creation, then creates a task with a required name and optional project link. The label constructor defines custom classes before the user uploads images or video. Advanced settings include image quality and segment size, which the speaker leaves aside for the basic example.
The annotation walkthrough explains bounding boxes and the distinction between shapes for single images and tracks for video objects across frames. It also covers polygons, polylines, points, ellipses, cuboids and masks, with examples of where each shape applies. The practical exercise concentrates on rectangular boxes around a worker and assigning predefined classes. Pressing N activates the drawing tool. The speaker recommends fitting boxes closely around objects to avoid excess background or clipped parts.
The final section exports the job as a dataset. It explains choosing a format for the intended training workflow, including COCO 1.0 and CVAT for images 1.1. Users can include images, set a custom filename and download a ZIP archive. This is an annotation and export guide; it does not demonstrate a self-hosted installation or model training.