labelme: a graphical image annotation tool with AI-assisted masking and multi-format dataset export

What it solves

Labelme is a graphical image annotation tool designed to help users create ground-truth data for computer vision tasks. It simplifies the process ofing labeling images for various AI models, such as those used for object detection and segmentation.

How it works

Written in Python and using the Qt framework for its interface, the tool allows users to manually draw shapes (polygons, rectangles, circles, lines, and points) over images to define objects. It also integrates AI-assisted features using SAM (Segment Anything Model), EfficientSAM, and YOLO-world for faster point-to-polygon/mask and text-to-annotation workflows. Annotations are saved as JSON files, which can then be exported to common dataset formats like VOC and COCO.

Who it’s for

It is intended for researchers and developers building computer vision models who need to create high-quality labeled datasets for image classification, bounding box detection, semantic segmentation, and instance segmentation.

Highlights

• Diverse Annotation Primitives: Supports polygon, rectangle, circle, line, and point tools.
• AI-Assisted Labeling: Integrates SAM, EfficientSAM, and YOLO-world for automated mask and text-based annotation.
• Multi-format Export: Exports data to VOC and COCO formats for semantic and instance segmentation.
• Video Annotation: Includes support for annotating video frames.
• Global Accessibility: Available in 20 different languages.
• Standalone App: Offers a standalone executable for users who do not want to manage Python or Qt dependencies.