ultralytics 8.2.64

v8.2.64 - `ultralytics 8.2.64` YOLOv10 SavedModel, TFlite, and GraphDef export (#14572)

on Python PyPI

🌟 Summary

The v8.2.64 release for Ultralytics provides critical updates including better token management for GitHub actions, enhancements in documentation for object detection models, expanded TensorFlow export support, and various compatibility improvements across the codebase.

📊 Key Changes

• GitHub Workflow Update: Improved secret token management in .github/workflows/format.yml.
• Documentation Update: Changes in the bounding box format within the datasets documentation, fixing markup issues.
• Model Export Enhancements: Expanded support for TensorFlow (TF SavedModel, TF GraphDef, TF Lite) formats.
• New Utility Functions: Introduction of torch_load in utility patches and autocast in torch_utils for better AMP compatibility.
• Miscellaneous Code Improvements: Adjustments in model loading for YOLO-NAS, TensorFlow export options, and improved interpolation handling in image transformation functions.

🎯 Purpose & Impact

• Better Token Management: Fixes the token usage in GitHub workflows to offer more robust and flexible secret management. This ensures smoother and more secure CI/CD flows.
• Enhanced Documentation: Improved bounding box format descriptions for more accurate and easy-to-understand object detection model training guidance. Makes it easier for users to train models with precise annotations.
• Broader Export Compatibility: Adding TensorFlow formats support means users can now export models in more formats, making it easier to integrate with a wider array of applications and tools.
• Utility Enhancements: The new torch_load function and autocast method offer greater flexibility and efficiency in model handling and automatic mixed precision (AMP) training, respectively, ensuring compatibility with both newer and older PyTorch versions.
• Overall Stability: Various bug fixes and improvements across the codebase improve the robustness, performance, and user-friendliness of the Ultralytics repository.

These updates further streamline the user experience, enhance compatibility across various platforms, and ensure the code adheres to the latest practices, significantly benefiting both expert developers and those newer to model training and deployment. 🚀