Refactor optimizer initialization

[glenn-jocher]

smart_optimizert() function now handles optimizer creation with smart bn and bias handling to avoid weight decay.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Simplification of optimizer configuration in YOLOv5 training script.

📊 Key Changes

• Removed specific optimizer configurations directly from train.py.
• Introduced smart_optimizer function in utils/torch_utils.py.
• The smart_optimizer function handles optimizer creation, consolidating parameter groups and optimizer setup.
• Support for RMSprop optimizer has been added.

🎯 Purpose & Impact

• Purpose:
  • Clean up the train.py script by removing redundant optimizer setup code.
  • Centralize the optimizer creation to improve maintainability and extend functionality.
• Impact:
  • Users can easily configure and switch between different optimizers, leading to potential improvements in training efficiency.
  • Provides a more streamlined and adaptable codebase for future updates or customizations.🛠