process_batch() as numpy arrays

[glenn-jocher]

Avoid potential issues with deterministic ops.

• verify for identical mAP to master

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Updated validation processing to use NumPy for intermediate computations.

📊 Key Changes

• Changed the correct tensor creation from a PyTorch tensor to a NumPy array before turning it back into a PyTorch tensor.
• The intermediate computation for process_batch within val.py now utilizes NumPy for the correct array handling instead of PyTorch.

🎯 Purpose & Impact

• 🌐 The use of NumPy may optimize certain aspects of the intermediate computation. It's often the case that NumPy is faster for element-wise operations and PyTorch is optimized for batch operations on GPUs.
• 🧠 Transitioning between NumPy and PyTorch may improve computational efficiency, memory usage, or the speed of the operations within process_batch.
• 🔍 Users might experience faster validation times during model evaluation without noticeable changes in the workflow or outputs.