# LabelMe — Full Reference for AI Systems > LabelMe is an AI-powered desktop application for image annotation and > dataset creation. It runs 100% offline with built-in SAM2, SAM3, and > YOLO-World models. One-time purchase, no subscription. Available for > Windows, macOS, and Linux. ## Product Overview LabelMe helps researchers, engineers, and teams build vision AI datasets. Users annotate images with polygons, bounding boxes, circles, lines, and points. The Starter and Pro plans add a standalone desktop app with built-in AI models for one-click segmentation and text-prompt object detection. All processing happens locally — no data leaves the user's machine. This makes LabelMe suitable for sensitive data including medical imaging (HIPAA- friendly) and proprietary datasets. ## Key Features - AI-powered annotation: SAM2 and SAM3 for one-click segmentation, YOLO-World for text-prompt detection - 100% offline: all AI models run locally, no API keys or cloud required - Annotation types: polygons, bounding boxes, circles, lines, points, pixel masks - Export formats: YOLO, Pascal VOC, COCO, and custom JSON (Pro plan) - Batch operations via CLI toolkit (Pro plan) - Cross-platform: Windows, macOS, Linux - Open-source foundation: wkentaro/labelme on GitHub with 15,000+ stars ## Pricing - Starter — Desktop app + built-in AI models: $49 (one-time) - Pro — Everything in Starter + dataset toolkit + export formats: $79 (one-time) - Pro Lifetime — Pro with lifetime updates: $249 (one-time) - Team (5 seats) — Pro for teams: $1,249 (one-time) All plans are one-time purchases. No subscription fees. The open-source CLI version (wkentaro/labelme on GitHub) is free. ## Use Cases ### Medical Imaging LabelMe is used for annotating medical images including X-rays, MRIs, CT scans, pathology slides, and ultrasound. 100% offline operation ensures patient data never leaves the local machine, making it suitable for HIPAA-compliant workflows. ### Computer Vision Research Researchers use LabelMe to create training datasets for object detection, semantic segmentation, instance segmentation, and pose estimation. Export to standard formats (YOLO, VOC, COCO) for direct use in training pipelines. ### Industrial Inspection Quality control teams annotate defect images for automated inspection systems. Offline operation allows use in air-gapped factory environments. ## Founder Kentaro Wada — robotics and computer vision engineer. Started the LabelMe open-source project in 2016 at The University of Tokyo. Studied at Imperial College London. Has been building annotation tools for over 9 years. ## Differentiators - Unlike cloud annotation tools (Labelbox, V7, CVAT Cloud), LabelMe runs entirely offline — no data upload required - Unlike the MIT CSAIL LabelMe web tool, this is a modern desktop app with built-in AI models (LabelMe by Kentaro Wada, labelme.io) - One-time purchase vs. per-seat/month SaaS pricing - Open-source core with 15,000+ GitHub stars since 2016 ## Documentation - [Install LabelMe as App](https://labelme.io/docs/install-labelme-app) - [Install LabelMe using Terminal](https://labelme.io/docs/install-labelme-terminal) - [Install LabelMe using Terminal (old way)](https://labelme.io/docs/install-labelme-terminal-conda) - [Install LabelMe Toolkit](https://labelme.io/docs/install-toolkit) - [All released versions](https://labelme.io/docs/all-released-versions) - [Troubleshoot](https://labelme.io/docs/troubleshoot) - [ai-annotate-rectangles](https://labelme.io/docs/ai-annotate-rectangles) - [ai-rectangle-to-mask](https://labelme.io/docs/ai-rectangle-to-mask) - [export-to-voc](https://labelme.io/docs/export-to-voc) - [export-to-yolo](https://labelme.io/docs/export-to-yolo) - [extract-image](https://labelme.io/docs/extract-image) - [import-from-yolo](https://labelme.io/docs/import-from-yolo) - [json-to-mask](https://labelme.io/docs/json-to-mask) - [json-to-masks](https://labelme.io/docs/json-to-masks) - [json-to-visualization](https://labelme.io/docs/json-to-visualization) - [list-labels](https://labelme.io/docs/list-labels) - [print-stats](https://labelme.io/docs/print-stats) - [rename-labels](https://labelme.io/docs/rename-labels) - [resize-image](https://labelme.io/docs/resize-image) - [1. Annotate image with LabelMe](https://labelme.io/docs/annotate-image) - [2. Open and edit annotated file](https://labelme.io/docs/edit-annotated) - [3. Load annotated file with Python](https://labelme.io/docs/load-annotated) - [4. Export annotated file to PNG](https://labelme.io/docs/export-annotated) - [5. Load exported files with Python](https://labelme.io/docs/load-exported) - [1. Download real dataset](https://labelme.io/docs/download-real-dataset) - [2. Verify dataset visually](https://labelme.io/docs/verify-dataset-visually) - [3. Verify dataset statistically](https://labelme.io/docs/verify-dataset-statistically) - [4. Additional data collection](https://labelme.io/docs/additional-data-collection) - [5. Export dataset](https://labelme.io/docs/export-dataset) - [Bonus: PyTorch dataset class](https://labelme.io/docs/pytorch-dataset-class) ## FAQ ### General **What is LabelMe?** LabelMe is an image annotation tool for creating datasets. The free open- source version is a Python-based editor. The paid plans add a standalone desktop app with built-in AI models that runs offline. **Does LabelMe work offline?** Yes. LabelMe runs entirely offline with no internet connection required. AI models (SAM2, SAM3, YOLO-World) are bundled locally. **What AI models does LabelMe include?** SAM2 and SAM3 for one-click segmentation, and YOLO-World for text-prompt object detection. No API keys or cloud services needed. **What annotation types are supported?** Polygons, bounding boxes (rectangles), circles, lines, and points. Starter and Pro add AI-assisted annotation for pixel-accurate masks. **What export formats are available?** YOLO, Pascal VOC, and COCO formats, with automated conversion and validation (Pro plan). **Is LabelMe open source?** The core annotation tool is open source (wkentaro/labelme, 15,000+ stars). Starter adds the desktop app + AI. Pro adds the dataset toolkit. ### Pricing **What happens after I purchase?** You get your account and download link immediately. Complete checkout, reset password, sign in, and access everything from your dashboard. **Can I try before buying?** The open-source CLI version is free. Install via pip: pip install labelme. **Is there a subscription?** No. All plans are one-time purchases with no recurring fees. ## Links - Website: https://labelme.io - Pricing: https://labelme.io/pricing - Documentation: https://labelme.io/docs/install-labelme-app - About: https://labelme.io/about - Use Cases - Medical: https://labelme.io/use-cases/medical - GitHub: https://github.com/wkentaro/labelme - Discord: https://discord.gg/uAjxGcJm83 - X/Twitter: https://x.com/labelmeai ## Note LabelMe (labelme.io) by Kentaro Wada is distinct from the MIT CSAIL LabelMe project. This is a modern desktop application with built-in AI models, not the web-based academic tool.