ultralytics 8.3.157

v8.3.157 - `ultralytics 8.3.157` Add `faster-coco-eval` package for COCO/LVIS evaluation (#17020)

on Python PyPI

🌟 Summary

This release introduces much faster and more modern COCO/LVIS evaluation by integrating the faster-coco-eval library, along with several improvements for segmentation, pose, dataset handling, documentation, and dependency management. 🚀

📊 Key Changes

• Major Evaluation Upgrade: Replaces the aging pycocotools with the high-speed faster-coco-eval library for COCO and LVIS dataset evaluation, resulting in up to 4.5x faster validation. ⚡
• Segmentation Improvements: Enhanced mask post-processing in YOLO segmentation for better compatibility and fewer errors with custom mask methods.
• Pose Model Reliability: Standardized keypoint data handling in YOLO pose models for more consistent and robust results.
• Dataset Flexibility: Improved label verification for grounding datasets—custom datasets now load without abrupt errors, while still validating known datasets.
• Documentation Enhancements:
  • Added a step-by-step YouTube tutorial for dog pose estimation with YOLO11. 🎥🐕
  • Updated GQA dataset links in YOLOE and YOLO-World docs for better clarity.
  • Added a direct button to the YOLO data augmentation guide in the Solutions docs.
• Dependency Updates:
  • Flask dependency is now pinned to version 3.0.1+ for similarity search, ensuring compatibility and security.
  • Updated urllib3 to 2.5.0 in RTDETR ONNXRuntime example.
  • OpenVINO export now requires version 2025.2.0+ on macOS 15.4+ for smoother exports.
• Other Fixes: Improved error handling when validation labels or predictions are missing.

🎯 Purpose & Impact

• Faster Validation: Dramatically reduces evaluation time for COCO/LVIS datasets, making model development and benchmarking much quicker for all users.
• Future-Proofing: Moves away from deprecated tools (pycocotools) to a modern, actively maintained alternative, ensuring long-term stability.
• Better User Experience: Reduces errors and confusion during segmentation, pose estimation, and dataset loading—especially for custom datasets.
• Improved Learning & Onboarding: Enhanced documentation and tutorials help both beginners and experienced users get started and optimize their workflows more easily.
• Increased Security & Compatibility: Dependency updates ensure users benefit from the latest features and security patches.

Overall, this update makes Ultralytics models faster, more reliable, and easier to use—whether you're training, validating, or just getting started! 🚀✨

What's Changed

• Fix predicted mask shape mismatch by @Laughing-q in #21083
• Update YOLOE and YOLOWorld GQA dataset link by @glenn-jocher in #21087
• Add https://youtu.be/J-RH22rwx1A to docs by @RizwanMunawar in #21103
• Add augmentation guide button by @RizwanMunawar in #21061
• Fix error when neither labels nor predictions exist during validation by @Y-T-G in #21038
• Bump urllib3 from 2.4.0 to 2.5.0 in /examples/RTDETR-ONNXRuntime-Python by @dependabot[bot] in #21110
• Pin flask>=3.0.1 for similarity search solution by @RizwanMunawar in #21112
• Enable verify_labels for GroundingDataset by @mohiuddin-khan-shiam in #21095
• Update macOS runners and pin OpenVINO>=2025.2.0 for macos-15 by @glenn-jocher in #20471
• Remove the unnecessary conditional check by @WillieMaddox in #21120
• ultralytics 8.3.157 Add faster-coco-eval package for COCO/LVIS evaluation by @MiXaiLL76 in #17020

New Contributors

• @WillieMaddox made their first contribution in #21120
• @MiXaiLL76 made their first contribution in #17020
• @mohiuddin-khan-shiam made their first contribution in #21095

Full Changelog: v8.3.156...v8.3.157