https://developer.apple.com/documentation/vision/recognizing_objects_in_live_capture
The app allows detecting objects in real-time by using compatible models*.
Optimized YOLOv5s model | Non-optimized YOLOv5s model |
---|---|
ANE optimized models ultralytics/yolov5#2526 will run at full speed. The structure of the model should have the following things:
- Supported layers, see reference for conversion of the YOLOv5 models https://github.com/danikkm/yolov5-coreml-tools
- Image, IoU, and confidence - input features
- Confidence and coordinates - output features
- Ability to select compute module (ANE, GPU+CPU, or CPU only)
- IoU and confidence thresholds adjustments
- Support for the front and back-facing cameras
- Ability to import .mlmodels from the Files app
- Add support for other camera modules (other than iPhone 11 and later)
- Add description view for the currently selected model by tapping on the label
- Rewrite settings view controller with native table view
- Implement edit actions in the table view, refactor