Attach transforms to model

[glenn-jocher]

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

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Refactored image classification training and validation scripts for consistency and simplicity.

📊 Key Changes

• 🔄 Moved the attachment of class names (model.names) to a more appropriate place in the training script to synchronize with the model information.
• ✂️ Removed unnecessary conversion of model to .float() in the validation file, aligning with the decision to remove half-precision inference (half=False).
• 🏷️ Standardized the use of model.names across both training and testing image displays for consistency.
• 🔗 Attached inference transforms (model.transforms) directly to the model within the training script for better accessibility.

🎯 Purpose & Impact

• 💡 Consistency: By moving model.names up to the model information section, it ensures class names are set before any logging or visualization, avoiding potential errors.
• 🔍 Clarity: Removing the float conversion from validation simplifies the code and the decision to go against half-precision inference ensures more stable and predictable model behavior.
• 🖼️ Visualization: Standardizing how class names are used in image displays helps maintain consistency during train and test visualization steps.
• 📈 Reliability: The changes prioritize model reliability and code maintainability without sacrificing performance, aiming for more robust and understandable image classification processes.