Apple MPS -> CPU NMS fallback strategy

[glenn-jocher]

Until more ops are fully supported this update will allow for seamless MPS inference (but slower MPS to CPU transfer before NMS, so slower NMS times).

python detect.py --device mps

Partially resolves #9596

Signed-off-by: Glenn Jocher [email protected]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhanced compatibility for Metal Performance Shaders (MPS) in non_max_suppression function.

📊 Key Changes

• Added a check to see if the MPS backend is being used.
• Converting prediction tensors to CPU if they are on an MPS device before Non-Maximum Suppression (NMS).

🎯 Purpose & Impact

• Purpose: To ensure the non_max_suppression function works even if the MPS backend is used, preventing any potential issues due to incomplete MPS support.
• Impact: Users working on Apple silicon Macs that leverage MPS for accelerated machine learning tasks will now experience smoother operation with YOLOv5, ensuring wider hardware compatibility. 🍏💻✨