Skip to content

Coming from LabelMe

Lannotex reads and writes LabelMe's JSON format, so you can bring an existing dataset over and still hand files back to LabelMe users.

Import a LabelMe folder

  1. Create a project (Combined pipeline enables every tool).
  2. Choose Project → Import LabelMe folder… and select the folder that contains your LabelMe .json files. Subfolders are included. Use Import annotations… for a single file.
  3. Lannotex reads each file, finds the image through its imagePath, checks that the dimensions match, and creates any missing labels.

Your LabelMe JSON files and images are left untouched.

What carries over

LabelMe Lannotex
rectangle shapes Bounding boxes
polygon shapes Polygons
circle shapes Polygons (32 points around the circle)
rotation (rotated rectangle) shapes Polygons with the four corners
flags set to true Image classes
Image files via imagePath Referenced in place, not copied

Current limitations

  • point, line, linestrip and mask shapes are skipped. The rest of each file still imports, and the import summary lists how many of each were skipped. Support for these shapes is planned.
  • Shapes with invalid geometry, such as a polygon with fewer than 3 points, are skipped and listed the same way.
  • imageData embedded in the JSON isn't used. The image file must exist on disk.
  • group_id, description, and per-shape flags aren't imported.

Export back to LabelMe

Export → Labelme JSON writes one LabelMe-compatible file per image. LabelMe can open these. Lannotex adds a lannotex block and a lannotex_id per shape, which LabelMe ignores and which let Lannotex re-import without creating duplicates.

How they compare

Both are good tools with different strengths. This comparison reflects LabelMe's published features and pricing as of October 2026.

Lannotex LabelMe
License MIT GPL-3.0 (source), EULA (paid desktop app)
Standalone desktop app Free Paid one-time license
YOLO export Free Pro tier
COCO export Yes Yes
Project storage One SQLite database per dataset One JSON file per image
Autosave & crash recovery Always on: recovery drafts + transactional saves Optional autosave
Validation before saving Geometry (bounds, self-intersection, zero area) and labels Label names
Team review with conflict handling Built in (LAN host) No
Headless export / validation lannotex export, lannotex analyze Conversion scripts in its examples
AI-assisted annotation (SAM, YOLO-World) Not yet Yes
Circles, lines, points, rotated boxes, masks Not yet Yes
Video annotation No Yes
UI languages English 20

If you depend on AI-assisted segmentation or the extra shape types today, LabelMe is the better fit for now. If you want a free, permissively licensed tool that keeps large datasets organized and safe, Lannotex is built for that.