ultralytics 8.3.175

v8.3.175 - `ultralytics 8.3.175` YOLOE `is_fused()` check prior to setting classes (#21605)

on Python PyPI

🌟 Summary

This release enhances model safety, improves documentation and logging, expands export testing, and updates example notebooks for a smoother and more reliable user experience. 🚦📚📝

📊 Key Changes

• Model Safety for YOLOE:
  • Prevents changing model classes after the model has been fused for inference, reducing risk of errors in production deployments.
• Improved YOLOE Documentation:
  • Added clear, step-by-step guides for fine-tuning and linear probing on both instance segmentation and object detection tasks.
• Enhanced Logging in Object Cropping Solutions:
  • Logs now show which object classes were detected and their counts, making results easier to interpret.
• Expanded Export Testing:
  • Added comprehensive tests for exporting YOLO models to NCNN format across various settings, ensuring greater reliability.
• Updated Example Notebooks:
  • All Colab and example notebooks now use the latest Ultralytics version, with refreshed outputs and improved user tips.
• Batch Size Handling Fix for TensorRT:
  • Validation and inference with TensorRT models now handle batch size more consistently and robustly.
• Queue Management Docs Improvement:
  • Added a Google Colab badge for instant access to the queue management notebook.

🎯 Purpose & Impact

• Safer Model Deployment:
  • Prevents accidental misconfiguration of YOLOE models after optimization, making deployments more robust and predictable.
• Easier Custom Training:
  • New documentation helps users fine-tune YOLOE models on their own datasets, whether for detection or segmentation, with less guesswork.
• Clearer Results & Debugging:
  • Enhanced logging and notebook updates make it easier to understand what the model is doing and to troubleshoot issues.
• Greater Export Reliability:
  • Expanded export tests catch more edge cases, leading to fewer surprises when deploying models in different environments.
• Improved Accessibility:
  • Quick-launch Colab badges and updated notebooks lower the barrier for new users to get started and experiment.

This update is recommended for all users who want safer workflows, better documentation, and a more seamless experience with Ultralytics models and tools. 🚀

What's Changed

• Add classes and total counts in solutions Logging by @RizwanMunawar in #21483
• Add NCNN matrix export tests to --slow CI by @lakshanthad in #21572
• Add queue-management notebook in docs by @RizwanMunawar in #21489
• Add YOLOE Detection fine-tuning example to Docs by @Y-T-G in #21530
• Update Google Colab notebooks by @glenn-jocher in #21583
• Fix validator incorrectly set batch_size for TensorRT inference by @syedhamzamohiuddin in #21592
• ultralytics 8.3.175 YOLOE is_fused() check prior to setting classes by @Y-T-G in #21605

New Contributors

• @syedhamzamohiuddin made their first contribution in #21592

Full Changelog: v8.3.174...v8.3.175