What it does

LabelImg is a desktop GUI for drawing bounding boxes around objects in images. It saves annotations in PASCAL VOC XML, YOLO text, or CreateML formats — the kind of grunt work that sits between raw photos and a trainable object-detection model. Python and Qt keep it cross-platform, with pip install, source builds, or even an X11 Docker container for the stubborn.

The interesting bit

The README’s top half is now a billboard for Label Studio, the multi-modal successor. The original tool still works — keyboard shortcuts, “difficult” flags, verification mode, the whole workflow — but active development stopped. It’s a rare case of a repo openly functioning as a redirect page while keeping the corpse on display.

Key highlights

• Outputs PASCAL VOC XML (ImageNet heritage), YOLO, and CreateML formats
• Keyboard-driven workflow: w for box, d/a for next/previous image, spacebar to verify
• “Difficult” flag for ambiguous objects that your model may want to skip
• Pre-defined class lists via plain text file
• Docker image available for dependency-averse users

Caveats

• No longer actively developed; README explicitly pushes users to Label Studio
• YOLO export discards the “difficult” flag and has restrictions on changing class lists mid-batch
• Qt/Python version friction is common enough that the docs recommend virtualenv specifically to avoid it

Verdict

Use it if you need a dead-simple offline box-drawer and don’t mind frozen code. Skip it if you want collaboration, non-image data, or any feature invented after roughly 2020 — the README will happily route you to Label Studio instead.