A YOLOv8 Desktop App That Pins Its Dependencies to the Wall

What it does

YoloSide puts a PySide6 graphical interface on top of Ultralytics YOLOv8 (and YOLOv5) inference. You drop a trained .pt model into the models folder, point the app at images or video, and view detections without touching a command line. Results are written to a local ./run directory.

The interesting bit

The project is essentially a Qt skin over the ultralytics Python package, but it is locked to an exact dependency snapshot—ultralytics==8.0.48 and pyside6==6.4.2. Deviate from either and the app throws a tuple-unpacking error. That rigidity makes it a time capsule of early-2023 YOLOv8 more than a living tool.

Key highlights

• Desktop GUI for running YOLOv8 and YOLOv5 inference
• Accepts custom models via .pt files placed in a models directory
• Detection results can be saved to a ./run output path
• UI built with Qt Designer (home.ui) and compiled to Python
• Author has discontinued maintenance and redirects users to active forks

Caveats

• Discontinued: the author explicitly states normal operation is not guaranteed
• Fragile dependencies: version drift in ultralytics or PySide6 causes a hard runtime error
• Several promised features—RTSP/camera input, target tracking, instance segmentation—remain on the todo list, and the author notes bugs remain

Verdict

Use it as a reference for building PySide6 frontends around YOLO, or harvest the UI layout for your own project. Avoid if you need maintained, stable object-detection software today.