What it does

JSON2YOLO converts COCO-format JSON annotations into the flat text files YOLO expects: one .txt per image, with normalized bounding boxes, segmentation masks, or keypoints. It also handles LabelMe polygons and Labelbox exports. The output follows the standard Ultralytics layout: data.yaml, images/, and labels/.

The interesting bit

The README’s “Important Update” is doing a lot of heavy lifting: the actual maintained code now lives in ultralytics.data.converter inside the main package. This repository is essentially a memorial plaque with some legacy scripts still attached. The standalone version hasn’t been abandoned entirely—existing workflows still run—but new development happens elsewhere.

Key highlights

• Supports COCO detection, segmentation, and keypoint annotations via convert_coco()
• LabelMe converter handles polygons, rectangles, circles, linestrips, and masks
• Labelbox bounding boxes convert; mask-only objects get skipped with a warning
• Output directory is deleted and recreated on each run (no incremental updates)
• Requires Python 3.8+

Caveats

• Standalone scripts in this repo are no longer actively updated
• --save-dir is wiped on every run, which is a footgun if you point it at the wrong directory
• Labelbox mask-only objects are silently skipped, not converted

Verdict

Use this if you’re maintaining an old pipeline that depends on the standalone scripts. Otherwise, pip install ultralytics and import convert_coco directly—it’s the same functionality with actual maintenance.