New Release: Ultralytics v8.3.149

:rocket: Ultralytics v8.3.149 Release: Enhanced Export, Smooth Video Handling, and Better Usability!

:glowing_star: Summary

We’re excited to announce Ultralytics v8.3.149, a release focused on export reliability, smoother video workflow integration, and expanded documentation to make your experience even better! This update brings significant quality-of-life improvements, especially for EdgeTPU and TensorFlow.js deployments, as well as video and visualization tasks.


:chart_increasing: New Features & Major Improvements

Export Fixes for EdgeTPU & TensorFlow.js

  • Resolved Model Export Bugs: Models with group convolutions now export smoothly to EdgeTPU and TensorFlow.js. Group convolutions are auto-disabled for these formats, eliminating previous errors and making your deployment process seamless.
    See PR by @Y-T-G

Improved Video Stream Handling

  • Refined Window and Video Management: OpenCV windows are now managed more reliably—no more unexpected closures or interruptions when working with video streams. Quitting streams is also easier and more predictable.
    See PR by @Y-T-G

Smarter Confusion Matrix Exports

  • Add Normalize & Precision Options: When exporting confusion matrices, you can now select normalization and control decimal places. This makes performance analysis and sharing more effective for both technical and non-technical audiences.
    See PR by @RizwanMunawar

Expanded Hardware Support

  • Clear IMX500 Export Coverage: Both YOLOv8n and YOLO11n models are now supported for Sony IMX500 export. Documentation and export tables have been updated for clarity and confidence.
    See PR by @RizwanMunawar

Documentation and Guide Upgrades

  • Updated Visualization Docs: The plot() and visualization argument docs are clearer and more accurate than ever, streamlining onboarding and reducing confusion.
    See PR by @RizwanMunawar
  • YOLOv7 ONNX/TensorRT Guide Improvements: The export and inference guide for YOLOv7 models has been revised for clarity, making export instructions and code samples easier to follow.
    See PR by @Laughing-q

:hammer_and_wrench: Other Notable Updates

  • Internal logic tweaks for more robust window and predictor handling
  • General doc and onboarding improvements throughout

Explore the full changelog for v8.3.149 for additional details and contributions from the community.


:bullseye: Why Upgrade?

  • Export effortlessly to more formats and devices—including EdgeTPU, TensorFlow.js, and Sony IMX500—without hitting architecture quirks.
  • Enjoy smoother video tasks with improved OpenCV window management across all inference and predictor workflows.
  • Gain sharper reporting with upgraded confusion matrix exporting—customize to fit your analysis and audience.
  • Benefit from clearer guides and documentation, making Ultralytics tools even more accessible for every skill level.

:speech_balloon: Ready to Try It?

Get started with Ultralytics v8.3.149 today!
Update via pip:

pip install -U ultralytics

We’d love to hear your feedback or questions—your experience helps drive continued improvements! Share your thoughts right here or open an Ultralytics Discussions topic.


A huge thank you to our community and team for making this release possible—your collaboration and input drive the YOLO ecosystem forward!

Release details: Ultralytics v8.3.149 on GitHub

Happy experimenting!