ultralytics/ultralytics v8.3.149

v8.3.149 - `ultralytics 8.3.149` Fix group-convolutions export bug for Edge TPU (#20919)

on GitHub

🌟 Summary

This release enhances model export compatibility, improves usability for video and visualization workflows, and expands documentation and reporting features across Ultralytics tools. 🚀

📊 Key Changes

• Export Fix for EdgeTPU & TF.js: Resolved a bug that caused errors when exporting models with group convolutions to EdgeTPU and TensorFlow.js formats. Group convolutions are now automatically disabled for these exports.
• Video Stream Handling Improvements: Refined how video streams and OpenCV windows are managed, preventing unwanted window closures and making quitting streams more reliable.
• Confusion Matrix Export Upgrades: Added options to normalize and set decimal precision when exporting confusion matrices, making results easier to analyze and share.
• Expanded IMX500 Export Support: Updated documentation and export tables to clarify that both YOLOv8n and YOLO11n models are now supported for IMX500 hardware.
• Visualization Documentation Cleanup: Improved and clarified documentation for visualization arguments, especially for the plot() method, ensuring all options are up-to-date and easy to find.
• YOLOv7 ONNX/TensorRT Guide Update: Made the YOLOv7 integration guide clearer and easier to follow, with improved code formatting and simplified export instructions.

🎯 Purpose & Impact

• Smoother EdgeTPU & TF.js Deployments: Users can now export models to EdgeTPU and TF.js without running into group convolution errors, making deployment on these platforms more reliable and accessible.
• Better Video & Window Management: Prevents accidental closure of unrelated OpenCV windows and reduces interruptions during video stream processing, improving the experience for users working with multiple video feeds.
• Flexible Reporting: Enhanced confusion matrix exports allow for more customizable and interpretable performance reports, supporting both technical and non-technical audiences.
• Clearer Hardware Support: Users working with Sony IMX500 hardware can confidently use the latest YOLO models, thanks to clearer documentation and expanded export support.
• Improved Usability & Onboarding: Updated guides and documentation make it easier for both new and experienced users to leverage Ultralytics features effectively, reducing confusion and setup time.

✨ This update is all about making Ultralytics tools more robust, user-friendly, and ready for deployment across a wider range of platforms and workflows!

What's Changed

• Enable normalize and decimals support for confusion matrix export by @RizwanMunawar in #20912
• Fix YOLOv7 ONNX/TensorRT inference guide by @Laughing-q in #20911
• Fix incorrect args in visualization-args.md by @RizwanMunawar in #20910
• Add yolo11n model support for IMX format in export table by @RizwanMunawar in #20884
• Move waitKey() and destroyAllWindows() to Predictor by @Y-T-G in #20913
• ultralytics 8.3.149 Fix group-convolutions export bug for Edge TPU by @Y-T-G in #20919

Full Changelog: v8.3.148...v8.3.149