New Release: Ultralytics v8.3.169

Discussion
, ,
1

:rocket: Ultralytics Release v8.3.169: Enhanced Visualizations & Smoother Workflow!

:glowing_star: Summary

We’re excited to announce Ultralytics v8.3.169! This release brings powerful new visualization features for model evaluation, plus enhancements for model exports, documentation, and overall reliability. Whether you’re debugging, training, or deploying, these updates make your workflow even smoother.

:framed_picture: New Features

Ground Truth & Prediction Visualizations

  • Confusion Matrix Visuals: Now easily visualize ground truths (GT), true positives (TP), false positives (FP), and false negatives (FN) directly from the confusion matrix during validation.
  • Multitask Support: These visualizations are available for Detect, Segment, Pose, and Oriented Bounding Box (OBB) models.
  • Simple Activation: Use the new visualize=True argument in your validation command to enable these powerful overlays and insights.

Huge thanks to @Y-T-G — see the feature PR #18868 for more details.

:open_book: Documentation Improvements

  • HomeObjects-3K Video Tutorial: A step-by-step YouTube tutorial demonstrates training and exporting with YOLO11. Perfect for newcomers and experienced users alike!
  • Clearer CLI & Export Instructions: Syntax issues fixed and guides improved for faster onboarding and fewer headaches (thanks @aaurelions, PR #21449).

:hammer_and_wrench: Improvements

  • Export & Namespacing: Exported models now handle class names more reliably, reducing errors during development and deployment (@Laughing-q, PR #21443).
  • RT-DETR Output Consistency: Inference predictions for RT-DETR models are now confidently sorted and respect max_det limits for predictable results (@atalaydenknalbant, PR #21457).
  • Streamlined TensorFlow Export: Unnecessary optimization flags have been removed for smoother, more compatible TensorFlow exports (@Y-T-G, PR #21482).

:lady_beetle: Bug Fixes

  • Fixed documentation examples and CLI syntax to reduce user confusion (@aaurelions, PR #21449).
  • Retaining predictor arguments when fetching model names improves workflow reliability (@Laughing-q, PR #21443).

:raising_hands: Community Contributions

Special thanks to our new contributors:

As always, the amazing progress in Ultralytics comes from the collaborative effort of the whole YOLO community and the Ultralytics team. We appreciate your continued support!

:link: Useful Links

:light_bulb: Try It Out & Share Feedback!

We invite you to upgrade to v8.3.169 and take the new visualization and workflow improvements for a spin. Your suggestions, bug reports, and ideas drive our progress. Please let us know your experiences—every bit of feedback helps make Ultralytics better for everyone.

Thank you for being part of the journey!

— The Ultralytics Team and YOLO Community