ultralytics/ultralytics v8.3.133

v8.3.133 - Raise RuntimeError when no valid images exist (#20606)

on GitHub

🌟 Summary

This release improves error handling for dataset loading, enhances NVIDIA Jetson support, and brings several usability and reliability updates to Ultralytics models. 🚀🛠️

📊 Key Changes

• Stricter Dataset Validation:
  • Now raises a clear error if no valid images are found in your dataset, instead of just showing a warning.
  • Improved warning messages when labels are missing or empty.
• NVIDIA Jetson Support:
  • Added official continuous integration (CI) testing for NVIDIA Jetson devices, ensuring better compatibility and reliability for edge AI users.
  • Adjusted GPU detection and test logic for Jetson-specific environments.
• Model & Export Usability:
  • Default image for OBB (Oriented Bounding Box) predictions is now a relevant sample ("boats.jpg") for easier out-of-the-box testing.
  • Added a helpful tip recommending OpenVINO export for Intel CPUs to improve performance.
• Metrics & Training Stability:
  • Improved model fitness calculation to safely handle missing or invalid metric values, preventing issues during training.
• User Guidance:
  • Updated error messages to reference more relevant dataset examples, making troubleshooting easier.

🎯 Purpose & Impact

• Faster Issue Detection:
  • Users are immediately alerted to critical dataset problems, saving time and preventing wasted training runs.
• Better Edge Device Support:
  • YOLO models are now more robust and reliable on NVIDIA Jetson hardware, benefiting developers deploying AI at the edge.
• Improved User Experience:
  • Clearer error messages and smarter defaults make it easier for both new and experienced users to get started and troubleshoot issues.
• Greater Training Reliability:
  • Enhanced metric handling and dataset checks reduce the risk of silent failures or confusing results during model development.

Overall, this update makes Ultralytics models more user-friendly, robust, and ready for a wider range of deployment scenarios. 🚀✨

What's Changed

• Prevent fitness from nan values by @Laughing-q in #20599
• Use boats.jpg for obb predict with Python code example by @RizwanMunawar in #20602
• Update error hint to reference coco8.yaml instead of dota8.yaml by @RizwanMunawar in #20592
• Add NVIDIA Jetson Self-Hosted GitHub Runners by @lakshanthad in #20438
• OpenVINO one-time message by @ambitious-octopus in #20538
• Raise RuntimeError when no valid images exist by @Y-T-G in #20606

Full Changelog: v8.3.132...v8.3.133