ultralytics 8.3.159

v8.3.159 - `ultralytics 8.3.159` Refactor and Clean up COCO evaluation (#21172)

on Python PyPI

🌟 Summary

This release refactors and unifies the COCO evaluation process across detection, segmentation, and pose models, streamlining the codebase and improving maintainability. It also brings enhancements to object counting, similarity search, NVIDIA Jetson benchmarking, and dependency management. 🛠️✨

📊 Key Changes

• Unified COCO Evaluation Logic: Introduced a single coco_evaluate method now shared by detection, segmentation, and pose validators, reducing code duplication and ensuring consistent metric reporting.
• Object Counter Display Fix: Improved logic so "IN" and "OUT" counts are only shown when their display options are enabled, making analytics cleaner and more intuitive.
• Similarity Search Refactor: The CLIP text model now supports direct image encoding, and the similarity search solution uses a unified model-loading approach for easier maintenance and extension.
• Validation Metrics Improvement: The save directory (save_dir) is now included in detection model validation metrics, making experiment tracking and result management easier.
• NVIDIA Jetson Documentation Update: Jetson YOLO11 benchmarks now use the larger COCO128 dataset and add support for the MNN model format, providing more reliable and up-to-date performance comparisons.
• Dependency Pinning for IMX: The Model Compression Toolkit version is now restricted to >=2.3.0 and <2.4.1 for IMX export and ONNX IMX inference, improving stability for Sony IMX workflows.
• Documentation & Link Updates: Numerous documentation links and references have been updated for accuracy, including tracker links and language selection menus.

🎯 Purpose & Impact

• Cleaner, More Maintainable Code: Centralizing evaluation logic reduces the risk of bugs and makes it easier to add new features or model types in the future. 🧹
• Consistent Model Evaluation: All model types now use the same robust evaluation process, ensuring fair and comparable results across detection, segmentation, and pose tasks.
• Improved User Experience: Object Counter analytics are clearer, and similarity search is more modular and user-friendly.
• Better Experiment Tracking: Including save_dir in metrics helps users organize and reproduce results more efficiently.
• More Reliable Jetson Benchmarks: Updated benchmarks and model format support provide users with more accurate performance data for edge deployments.
• Greater Stability for IMX Users: Pinning dependencies prevents compatibility issues, ensuring smoother exports and inference on Sony IMX hardware.
• Enhanced Documentation: Updated links and clearer instructions make it easier for users to find resources and avoid confusion.

Overall, this release focuses on making the Ultralytics ecosystem more robust, user-friendly, and future-proof for both developers and end users. 🚀

What's Changed

• Fix count display check for show_in=False and show_out=False by @fn-hide in #21047
• Add save_dir to Metrics for better access by @RizwanMunawar in #21136
• Update NVIDIA Jetson Doc with COCO128 Benchmarks by @lakshanthad in #21143
• Update 301 redirects by @glenn-jocher in #21148
• Use TextModel class for similarity search by @RizwanMunawar in #21114
• ultralytics 8.3.159 pin model-compression-toolkit>=2.3.0,<2.4.1 by @Laughing-q in #21161
• ultralytics 8.3.159 Refactor and Clean up COCO evaluation by @Laughing-q in #21172

New Contributors

• @fn-hide made their first contribution in #21047

Full Changelog: v8.3.158...v8.3.159