wkentaro/labelme

What it solves

Labelme provides a graphical interface for annotating images and videos to create training data for machine learning models. It simplifies the process of labeling objects for tasks like classification, bounding box detection, and semantic or instance segmentation.

How it works

Users open images or folders of images in the Labelme GUI, where they can draw various shapes—such as polygons, rectangles, circles, lines, or points—to define objects. The tool saves these annotations as JSON files. It also integrates AI-assisted features, allowing users to generate annotations automatically using models like SAM (Segment Anything Model) or YOLO-world. The project supports exporting data into standard formats like VOC and COCO.

Who it’s for

This tool is designed for researchers, data scientists, and developers who need to manually label or semi-automate the creation of high-quality datasets for computer vision projects.

Highlights

• Supports multiple annotation types: polygons, rectangles, circles, lines, and points.
• Includes AI-assisted labeling using SAM, EfficientSAM, and YOLO-world models.
• Exports data in standard formats compatible with common computer vision workflows (VOC, COCO).
• Available as a cross-platform desktop application or a Python package.
• Supports video annotation and multi-language interfaces.