ContextQMD
Back to Ultralytics

LabelMe to YOLO Dataset Export - Ultralytics Platform

docs/en/platform/integrations/labelme.md

8.4.1165.2 KB
Original Source

Export LabelMe Annotations to YOLO and Ultralytics Platform

LabelMe is an offline image annotation tool with an open-source Python application. There is no live LabelMe connection or API key to configure in Ultralytics Platform. The complete integration is a local workflow: annotate in LabelMe, convert the LabelMe JSON annotations to YOLO format with the LabelMe Toolkit, and upload the resulting ZIP as a Platform dataset.

1. Annotate the Images in LabelMe

Install LabelMe using the desktop app or the open-source Python package, then open the directory containing your images. LabelMe saves each image's annotations in a matching JSON file.

For the YOLO detection workflow in this guide, draw rectangles around each object and assign a class name. The LabelMe starter guide covers opening images, drawing shapes, and saving annotations, while the LabelMe dataset guide covers reviewing and preparing a complete annotated dataset.

Your source directory should contain the images and LabelMe JSON files:

text
your_dataset/
├── image_001.jpg
├── image_001.json
├── image_002.jpg
└── image_002.json

2. Install the LabelMe Toolkit

Install the LabelMe Toolkit by following LabelMe's toolkit installation guide. The toolkit and its exports run locally.

!!! info "LabelMe Pro is required for the export"

`export-to-yolo` is part of the LabelMe Pro Toolkit, and downloading its installer requires a LabelMe sign-in.
This is a LabelMe product requirement; the resulting ZIP can be uploaded on any Platform plan.

Verify the installation, then list every label found in the source dataset:

bash
labelmetk --version
labelmetk list-labels your_dataset/

Review the output before exporting. Labels omitted from --class-names are skipped, and the order you provide becomes the YOLO class ID order.

3. Export to YOLO Format

Run export-to-yolo with the source directory and a comma-separated list of class names:

bash
labelmetk export-to-yolo your_dataset/ --class-names crack,normal

Replace crack,normal with the labels returned by list-labels. LabelMe writes the result to your_dataset.export/:

text
your_dataset.export/
├── classes.txt
├── images/
│   ├── image_001.jpg
│   └── image_002.jpg
└── labels/
    ├── image_001.txt
    └── image_002.txt

classes.txt preserves the class names in the same order used by the YOLO label files. Keep it at the root of the export.

!!! note "This workflow creates a detection dataset"

LabelMe Toolkit exports rectangles as YOLO bounding boxes. It also reduces polygons and masks to their
axis-aligned bounding boxes, so `export-to-yolo` does not preserve segmentation geometry. Draw rectangles when
preparing a detection dataset with this workflow.

4. Create the ZIP Archive

Compress the contents of your_dataset.export/, not the directory around them. On macOS or Linux:

bash
cd your_dataset.export
zip -r ../your_dataset.zip classes.txt images labels

On Windows PowerShell:

powershell
Compress-Archive -Path .\your_dataset.export\* -DestinationPath .\your_dataset.zip

Open the ZIP before uploading and confirm that classes.txt, images/, and labels/ are at its root. An extra your_dataset.export/ wrapper prevents Platform from finding the root class list.

5. Upload to Ultralytics Platform

  1. Open Settings > Integrations > LabelMe.
  2. Click Upload export.
  3. Upload your_dataset.zip and finish creating the dataset.
  4. Wait for processing to complete, then review the images, classes, and annotations.
  5. Edit the annotations, train a model, and deploy it from the same workspace.

<!-- screenshot -->

LabelMe and the YOLO export remain entirely offline. Only the ZIP file you select in the upload dialog is sent to Platform.

Troubleshooting

  • Classes are named class0, class1, and so on: confirm that classes.txt is present at the root of the ZIP.
  • Some annotations are missing: run labelmetk list-labels your_dataset/ again and include every required label in --class-names.
  • Polygons became boxes: this is the documented behavior of LabelMe's export-to-yolo; it converts non-rectangle shapes to bounding boxes.
  • The dataset has no images: confirm that the exported images/ directory is included in the ZIP.
  • Platform cannot find the classes: remove any outer directory from the archive so classes.txt, images/, and labels/ are the top-level entries.

For the complete local workflow, see LabelMe's YOLO training guide and export-to-yolo reference.