labelme

Why we included this project

Labelme is the tool most teams reach for when they want to build their own computer vision dataset instead of paying for a hosted labeling service. It is a Qt-based graphical editor: you draw polygons, rectangles, circles, lines, or points over an image, and it exports the annotations in VOC and COCO formats that segmentation, detection, and classification models can consume directly. The AI-assisted modes, built around models like SAM and YOLO-world, generate masks or annotations from a single point or a text prompt, which removes the most tedious part of labeling thousands of images. Since it is a plain Python package with a documented JSON annotation format, you can script it and extend it, or drop it into your own data-prep pipeline. If you are putting together a custom dataset for a vision model, this is a solid, battle-tested place to start.