Update DetectMultiBackend for tuple outputs 2

[glenn-jocher]

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

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improved model inference and validation by streamlining output handling.

📊 Key Changes

• 🔨 Modified forward method in common.py to handle outputs more effectively.
• 🚮 Removed val parameter from forward method to simplify the calling interface.
• ➕ Added from_numpy method to convert NumPy arrays to PyTorch tensors only when necessary.
• ✨ Updated non_max_suppression in general.py to accept list/tuple inputs, catering to different model output types.
• 🧹 Cleaned up inference and loss computation in val.py by eliminating unnecessary conditionals.

🎯 Purpose & Impact

• 🚀 Streamlining the output processing enhances code maintainability and reduces potential for bugs.
• 📈 The removal of superfluous parameters leads to clearer function calls, making the API easier for developers to use.
• 🧠 The impact on users includes potentially faster inference times and smoother integration when using the model for different purposes, such as training or validation.