Labelme - The offline image annotation for vision AI

Holiday season sale 20% off (ends 01/07)

The offline image annotation for vision AI

Join 15,000+ builders who use Labelme to create training data for vision AI securely. No cloud uploads. 100% Private.

Efficient annotation with simple & intuitive UI
Fast annotation with AI auto-labeling and smart suggestions
On-device app, tools, and AI models keeping your data private

Get Labelme — $49

See How It Works

or download 3-day free trial version

How it works (workflow)

Annotate images with polygons, rectangles, and more. Validate & export annotations as a dataset.

Products (what we provide)

AI-powered annotation app, automation-ready tools & guides to build high-quality datasets.

Examples (what you can do)

Self-driving, retail, agriculture, medical, and more. Build datasets for various computer vision tasks.

Story (how we got here)

Automation obsessor. Solo-built Labelme (15K+ stars), GH 9+ Top trending. PhD, Lead CV at Mujin.

How it works

1. Annotate images with polygon, rectangle, or any shapes with Annotation app

2. Validate & export annotations as a dataset with Tookit

3. Train your AI model

How Annotation app works

Easy-to-use UI

Everything you need in a single window.

Offline & private

Keep data private & work confidently.

Any shapes

Polygon, rectangle, circle, line, & more.

Rich attributes

Add details (flags, texts) to annotations.

Annotation app

Offline-first app you download and run on your computer.

How Toolkit works (automation)

Batch process

Annotate, edit, validate, and export many at once.

The toolkit streamlines dataset creation and complements manual annotation.

Automation ready

Integrate with your workflow and other tools.

The app and toolkit are accessible from CLI, easily connectable.

Toolkit

Command-line tools to streamline dataset creation.

How AI works (annotation app & toolkit)

Prompt to annotations

Type what to annotate, to get it annotated with AI.

AI annotates objects with rectangles that best match your text prompt.

A click to annotation

Instant object annotation by a few clicks with AI.

You'll get real-time feedback. Tweak clicks on the fly, or the polygon later.

Rectangle to polygon

Convert rough rectangles to accurate polygons with AI.

Perfect for objects with complex shapes (e.g., irregular, curved).

Products

Annotation app

Desktop app to annotate images with polygons, rectangles, and more.

Preview

Starter guide

Guide to get you started with annotating the first image fast.

Preview

Toolkit

A set of command-line tools to streamline dataset creation.

Preview

Dataset guide

Guide to get you create, validate, and export a high-quality dataset fast.

Preview

Features table

Feature	How it helps you	Starter	Pro
Manual annotation (Annotation app)
Easy installation	Get started quickly (Windows, macOS, Linux)
Portable exe for Windows	No installation needed (e.g., use on work computer)
Offline annotation	No privacy concerns
Polygon annotation	For arbitrary shapes
Rectangle annotation	For bounding box detection
Point / Line / LineStrip / Circle annotation	For various use cases
Grouping of annotations	Annotate composite / multi-part objects
Text & flag attachment to annotations	Detailed metadata annotation
AI-assisted point to polygon / mask	Fast complex shape annotation
AI-assisted text prompt to rectangle	Fast rectangle annotation
Automation (Toolkit)
Export YOLO format	Train YOLO models
Import YOLO format	Import existing annotations for YOLO
Export VOC format	Train with common frameworks
Export mask-JPEGs	For segmentation training (multi-class)
AI text prompt to rectangle (batch)	Auto-annotate many images at once
AI rectangle to mask	Auto-convert rectangles to masks
Show statistics of dataset	Understand dataset composition
Rename labels in dataset	Standardize naming, fix typos
Resize images and annotations	Reduce image size to save disk space
More automations	Many more to help your workflow
Feature	How it helps you	Starter	Pro

Examples

Make image annotations for any applications with any shapes & AI-assist.

Self-driving car

(car as rectangle, lane as line)

Natural photo

(person, sofa as polygon)

Similar shaped objects

(fruit as polygon, duplicate for fast annotation)

A click to annotate by AI

(click a dog to annotate as polygon)

Text prompt to annotation by AI

(prompt 'person, shoe, bus, ...')

Object with holes

(group 'wheel' and '_background_')

Image classification (image-level flag)

cat: true/false, dog: true/false

Brightness/Contrast adjustment

(adjust without affecting original image)

Cartoon image: face detection by AI

(auto-annotate as rectangle by prompt 'face')

What they say

GitHub stars

15K+

App reviews

130+

Story

Kentaro Wada

Founder

Automation obsessor. Solo-built Labelme (15K+ stars), Gdown (4.7K+), and 9+ times top trending developers on GitHub.

Lead computer vision at Mujin (robotics startup), PhD at Imperial (early graduate, full-funded), UTokyo JSK robotics lab (first who went abroad in its 50-year history).

GitHub (23K+ stars, 4.8K+ forks)Twitter (1K+ followers)LinkedIn

Hi, I'm Kentaro!

Here is the story behind Labelme.

I built it when I was at the end of my bachelor's degree / beginning of my master's. I was working in a robotics lab and created it to teach my robot to see the world.

Around 2015-2016, Amazon Robotics held the Amazon Picking Challenge (APC) shortly after acquiring Kiva Robotics, the company who developed the AGVs used at Amazon. They had automated material delivery in warehouses but were looking for an automation solution for material picking.

We participated as a team from the lab (Can you spot me in the picture?), and I ended up building Labelme for the challenge.

For the full story, check out my tweet.