labelme-to-yolo 0.2.1

A tool to convert LabelMe dataset annotations into YOLO format for instance segmentation.

labelme-to-yolo

Convert LabelMe polygon annotations to Ultralytics YOLO format for instance segmentation — one command, ready-to-train dataset.

This project is the active continuation of the archived labelme2yolov7segmentation repository. All future development happens here.

---

The Problem

You annotated your dataset with LabelMe — drawing polygons, assigning labels, exporting JSON files. Now you want to train a YOLO model and you discover that YOLO expects a completely different format: normalized coordinates, one .txt per image, a specific folder structure, and a project.yml config file.

Converting by hand is tedious and error-prone. labelme-to-yolo does it in a single command.

---

What it does

• Reads all LabelMe .json annotation files from a folder
• Normalizes polygon coordinates to the [0, 1] range expected by YOLO
• Copies images and writes .txt label files into the YOLO folder structure
• Automatically splits your dataset into train / val / test sets
• Generates the project.yml configuration file ready to pass to Ultralytics

---

Quickstart

Install from PyPI:

pip install labelme-to-yolo

No install required (via pipx):

pipx run labelme-to-yolo --source-path /labelme/dataset --output-path /yolo/dataset

Run the conversion:

labelme2yolo --source-path /labelme/dataset --output-path /yolo/dataset

---

Usage

labelme2yolo --source-path PATH --output-path PATH

Option	Description
--source-path	Folder containing LabelMe .json files and their matching images
--output-path	Destination folder for the YOLO dataset

Both relative and absolute paths are supported.

Expected output

Running:

labelme2yolo --source-path /labelme/dataset --output-path /yolo/datasets

Produces:

datasets/
├── images/
│   ├── train/
│   │   ├── img_1.jpg
│   │   └── img_2.jpg
│   ├── val/
│   │   └── img_3.jpg
│   └── test/
│       └── img_4.jpg
├── labels/
│   ├── train/
│   │   ├── img_1.txt
│   │   └── img_2.txt
│   ├── val/
│   │   └── img_3.txt
│   └── test/
│       └── img_4.txt
├── train.txt
├── val.txt
├── test.txt
└── project.yml

The generated project.yml can be passed directly to Ultralytics YOLO:

from ultralytics import YOLO

model = YOLO("yolov8n-seg.pt")
model.train(data="/yolo/datasets/project.yml", epochs=100)

---

Installation from source

git clone https://github.com/Tlaloc-Es/labelme-to-yolo.git
cd labelme-to-yolo
pip install -e .

---

Contributing

Contributions are welcome.

1. Fork the repository
2. Create a branch: git checkout -b my-feature
3. Commit your changes following Conventional Commits
4. Run the tests: uv run poe test
5. Push your branch and open a pull request

---

⭐ If this saved you time, a star helps others find it

Stars help labelme-to-yolo appear when developers search for LabelMe and YOLO conversion tools. It takes 2 seconds.

⭐ Star on GitHub

---

License

MIT. See LICENSE.

Donation

If you want to support the project you can make a donation at https://www.buymeacoffee.com/tlaloc — thanks in advance.