Fix bugs

[Lornatang]

fix TypeError: non_max_suppression() got an unexpected keyword argument 'fast'

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Introducing YOLOv4 configuration and updating dependencies for consistency.

📊 Key Changes

• Updated non_max_suppression call in detect.py to use merge=True instead of fast=True.
• Added yolov4.yaml file, defining the configuration for the YOLOv4 model.
• Pinned specific versions for torch and torchvision in requirements.txt to ensure compatibility.

🎯 Purpose & Impact

• The change in non-max suppression improves the accuracy of bounding box predictions by merging overlapping boxes 📦.
• The new YOLOv4 model configuration enables users to train and use a more advanced, state-of-the-art model for object detection 🤖.
• Specifying exact versions of torch and torchvision reduces issues due to API changes in these libraries, making the setup process more reliable for users 🔧.

[glenn-jocher]

@Lornatang thanks! I can't merge this because of the dependency issues, we can't force 1.4.

I've pushed a fix for the detect.py bug though, thank you for noticing this!

[Lornatang]

Ok.