anylabeling: an AI-assisted image annotation tool with auto-labeling via YOLOv8 and Segment Anything

What it solves

AnyLabeling is an image annotation tool designed to reduce the manual effort of labeling data for computer vision tasks. It combines the functionality of LabelImg and Labelme with an improved user interface and AI-powered auto-labeling capabilities to speed up the creation of training datasets.

How it works

The tool provides a graphical interface for creating polygons, rectangles, circles, lines, and points. To accelerate the process, it integrates several AI models for automatic annotation:

• Segment Anything (SAM) Family: Supports SAM (ViT-B/L/H), MobileSAM, SAM 2, SAM 2.1, and SAM 3. These models allow users to segment objects using point, rectangle, or text prompts (SAM 3).
• YOLOv8: Used for automatic object detection and labeling.
• OCR/KIE: Includes specialized labeling for text detection, recognition, and Key Information Extraction.

Who it’s for

It is intended for data scientists, ML engineers, and data labelers who need to create high-quality ground-truth data for object detection, instance segmentation, OCR, and pose estimation projects.

Highlights

• AI-Assisted Labeling: Integrates YOLOv8 and the entire Segment Anything family for rapid auto-labeling.
• Crosspatform Support: Available as a standalone executable or via PyPI for Windows, macOS, and Linux.
• Crosspatform Support: Supports multiple languages including English, Vietnamese, and Chinese.
• Versatile Annotation: Supports a wide range of shapes (polygons, rectangles, circles, lines, points) and applications like medical imaging and 2D pose estimation.