val.py --weights and --data compatibility check

[glenn-jocher]

Improved error messages for understanding of user error with val.py. May help #7291

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhanced model-dataset compatibility verification in validation script.

📊 Key Changes

• Added a check to ensure the number of classes (nc) matches between the loaded weights (--weights) and the dataset configuration (--data) during non-training runs.

🎯 Purpose & Impact

• 🛠 Purpose: Prevents potential mismatch errors by verifying that the model weights correspond to the expected number of classes defined in the dataset being used for validation.
• 🔍 Impact: Increases user confidence that the model is being evaluated correctly, ensuring that validation results are reliable and consistent with the training setup.

[glenn-jocher]

Example result:

$ python val.py --weights yolov5s.pt --data voc.yaml
...
AssertionError: yolov5s.pt (80 classes) trained on different --data than what you passed (20 classes). Pass correct combination of --weights and --data that are trained together.