Support YOLOv5

[tucan9389]

No description provided.

[tucan9389]

Convert pre-trained YOLOv5 pt to Core ML

Clone and install pre-requirements

Note that you should clone mshamash/yolov5 repo and checkout fix/coreml_export_nms_layer branch, instead of the original repo (ultralytics/yolov5) because the PR wasn't merged.

git clone https://github.com/mshamash/yolov5
cd yolov5
git checkout fix/coreml_export_nms_layer
pip install -qr requirements.txt  # install

Export to Core ML model

python export.py --weights yolov5n.pt --include coreml
python export.py --weights yolov5s.pt --include coreml
python export.py --weights yolov5m.pt --include coreml
python export.py --weights yolov5l.pt --include coreml
python export.py --weights yolov5x.pt --include coreml

python export.py --weights yolov5n6.pt --include coreml
python export.py --weights yolov5s6.pt --include coreml
python export.py --weights yolov5m6.pt --include coreml
python export.py --weights yolov5l6.pt --include coreml
python export.py --weights yolov5x6.pt --include coreml

Check the model input/output with https://netron.app

yolov5s.mlmodel example

Check the model with Xcode

Related Issue and PR

• Add NMS to CoreML model output, works with Vision ultralytics/yolov5#7263
• when exporting to coreml (mlmodel format) NMS is not added. ultralytics/yolov5#7011
• Introduce iOS demo repo for YOLOv5 Core ML models 📱 ultralytics/yolov5#9271

[tucan9389]

Class names of YOLOv5's pre-trained with COCO dataset

• https://github.com/ultralytics/yolov5/blob/9da6d0f9f5bc37fa386b7b82d2a963f94650949a/data/coco.yaml#L17-L97
• 80 classes