← all repositories
developer0hye/Yolo_Label

A YOLO labeler that admits labeling is hell

A Qt-based desktop app for drawing bounding boxes, with a keyboard-heavy workflow and optional ONNX auto-labeling to speed up the misery.

Yolo_Label
Velocity · 7d
+0.2
★ / day
Trend
steady
star history

What it does

YOLO-Label is a cross-platform Qt 6 desktop app for annotating object-detection datasets in YOLO format. You load a folder of images and a class-names file, then draw bounding boxes. It ships as pre-built binaries for Windows, Linux (AppImage), and macOS (Apple Silicon), or you can compile from source.

The interesting bit

The author openly hates manual labeling and designed around that. Instead of drag-and-drop, you draw boxes with two left-clicks (corner to corner), which they claim reduces wrist strain. The keyboard shortcuts are extensive: arrow keys nudge or resize boxes, R triggers auto-labeling from a local ONNX model, and mouse-wheel over the image navigates forward/back. There’s even a focus-aware timer in the status bar tracking how many hours you’ve spent in annotation purgatory.

Auto-labeling is the real time-saver. Load any Ultralytics-exported YOLO model (v5 through v12, end-to-end with NMS baked in) and it reads class names, input size, and config from the ONNX metadata automatically. No separate config files. There’s also a cloud fallback via yololabel.com for open-vocabulary detection if you lack a local GPU.

Key highlights

  • Two-click box drawing instead of drag-and-drop; keyboard nudge/resize with 1px and 5px steps
  • Local ONNX inference for auto-labeling, supporting YOLOv5 through YOLOv12 with automatic metadata extraction
  • Cloud auto-labeling integration (yololabel.com) with batch processing up to 20 images per request
  • Copy/paste bounding boxes between images, undo/redo, and real-time contrast adjustment
  • Cross-platform: Windows x64, Linux AppImage, macOS Apple Silicon

Caveats

  • The “sensitive” branding is mostly about UI responsiveness and keyboard shortcuts; the README never quantifies what “sensitive” means technically
  • Cloud auto-labeling requires creating an account and API key on a third-party service
  • Build-from-source with ONNX support requires a separate script to download ONNX Runtime first

Verdict

Worth a look if you’re labeling YOLO datasets regularly and find existing tools sluggish or wrist-unfriendly. Skip it if you need team collaboration, video annotation, or non-YOLO export formats — this is a solo desktop tool with a narrow scope.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.