New Release: Ultralytics v8.3.133

Discussion
,
1

:rocket: Ultralytics v8.3.133 Release Announcement

:glowing_star: Summary

We’re excited to announce the release of Ultralytics v8.3.133! This update brings improved error handling for datasets, enhanced support for NVIDIA Jetson devices, usability upgrades for model export and testing, and several reliability improvements that make working with Ultralytics models smoother and more robust than ever.

:wrapped_gift: New Features & Enhancements

Stricter Dataset Validation

  • Get immediate, clear feedback if your dataset contains no valid images—preventing wasted training runs and making debugging easier.
  • Improved warnings for missing or empty labels help ensure your data quality is top-notch.
    PR by @Y-T-G | PR by @RizwanMunawar

NVIDIA Jetson Support

  • Official CI testing on NVIDIA Jetson devices: YOLO models are now easier to deploy and more reliable on edge AI hardware.
  • Smarter GPU detection and test logic tailored for Jetson-specific environments.
    PR by @lakshanthad

Model & Export Usability

  • Oriented Bounding Box (OBB) predictions now default to a relevant sample image, "boats.jpg", for an easier and more intuitive experience right out of the box.
  • When exporting models for Intel CPUs, you’ll now see a helpful tip recommending OpenVINO export for improved performance.
    PR by @RizwanMunawar | PR by @ambitious-octopus

:wrench: Improvements

  • Metrics & Training Stability:
    Model fitness calculations now safely handle missing or invalid metrics, reducing the risk of silent failures or ambiguous training results.
    PR by @Laughing-q

  • User Guidance Updates:
    Error messages and hints are now more relevant and direct users to helpful example datasets for quicker troubleshooting.
    PR by @RizwanMunawar

:lady_beetle: Bug Fixes

  • Prevented fitness from producing nan (not-a-number) values by safely handling invalid or missing metric values during model evaluation.
  • Improved error handling and default behaviors to remove ambiguity and confusion for end users.

:bullseye: Why This Matters

  • Save Time: Early and clear detection of data problems prevents wasted resources and increases productivity.
  • Edge Deployment Ready: Jetson support ensures YOLO models are robust for edge AI applications.
  • User-Friendly Experience: Smarter defaults and clearer messages help both new users and seasoned pros get models running quickly and troubleshoot with confidence.
  • Reliable Training: Enhanced metric handling and data validation keeps your development on track.

:memo: How to Try & Share Feedback

A huge thank you to everyone contributing ideas, bug reports, and pull requests to Ultralytics. Every step forward is made possible by our incredible YOLO community and teamwork. We look forward to hearing how the new release works for you! :rocket: