Clip Objects365 autodownload labels

[glenn-jocher]

Fixes out of bounds labels that seem to affect ~10% of images in dataset.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Upgraded bounding box calculations for Objects365 dataset configuration in YOLOv5.

📊 Key Changes

• Imported additional utility functions, including xyxy2xywhn from utils.general.
• Revised the way bounding box dimensions are calculated, now using xyxy2xywhn for normalization and clipping.

🎯 Purpose & Impact

• 🎨 Improved Accuracy: Using xyxy2xywhn helps ensure bounding box coordinates are normalized and clipped correctly, which may reduce errors during training and inference.
• 🛠️ Code Maintainability: Relying on a dedicated utility function for coordinate transformations makes the codebase cleaner and more maintainable.
• 🔍 Data Quality: Ensures that the bounding box data for the Objects365 dataset is more precise, potentially leading to improved model performance for users working with this dataset.

[glenn-jocher]

Object365 trains with 0 corrupt images with this PR :)