New Release: Ultralytics v8.3.68

Discussion
,
1

:rocket: New Release: Ultralytics v8.3.68

We are thrilled to announce the release of Ultralytics YOLOv8.3.68, packed with valuable updates to stay at the forefront of AI innovation. This release focuses on improving benchmarking workflows, enhancing export robustness, refining documentation, and providing advanced model comparison tools. Whether you’re training, exporting, or benchmarking your models, this release ensures greater ease and reliability for your tasks! :tada::sparkles:

:star2: Summary

  • Enhanced benchmarking tools and log clarity
  • EfficientDet (d0-d3) model integration for comparison
  • Resolved edge-case export issues for ONNX, OpenVINO, and TFLite
  • Updated documentation for clarity and developer usability

:mag: Key Features

New Features

  • Added EfficientDet (d0-d3) to benchmarking for richer model performance analysis.
    (See PR by @glenn-jocher)
  • Introduced Javascript-based active model configuration for better visualization during benchmarking.
    (See PR by @glenn-jocher)

Improvements

  • Simplified benchmark chart legends for cleaner visualization.
    (See PR by @glenn-jocher)
  • Enhanced handling of model paths in benchmarking for more robust log clarity and fallbacks.
    (See PR by @Y-T-G)
  • Adjusted export configurations, fixing issues with ONNX, OpenVINO int8, and TFLite edge cases (imgsz=32).
    (See PR by @Y-T-G)
  • Improved dataset logic for benchmarks and export process automation. (See PR by @glenn-jocher)

Documentation Updates

  • Updated AzureML quickstart recommendations for Python versions.
    (See PR by @Lucashygi)
  • Added a fallback mechanism to ensure documentation builds even with file minification issues.
    (See PR by @glenn-jocher)

:dart: Impact

These updates provide greater flexibility, reliability, and usability to users:

  • Clearer logs and benchmarking outputs make debugging and evaluation easier.
  • Enhanced flexibility for export tasks, ensuring models work efficiently for diverse setups.
  • Improved user experience for both advanced developers and newcomers by clarifying setup and documentation.
  • Smart benchmarking tools to help compare models (such as EfficientDet) against YOLO architectures.

:man_technologist: What’s Changed

  • Simplify benchmark chart legend (#18878)
  • EfficientDet model benchmarking added (#18884)
  • Add Javascript active model argument (#18886)
  • Fallback mechanism for documentation minification (#18887)
  • Fixes to TFLite and OpenVINO export edge cases (#18898)
  • Benchmarking robustness improvements (#18894)

Full Changelog: v8.3.68 Release Notes

:tada: A Big Welcome to New Contributors

  • @Lucashygi made their first contribution with AzureML Quickstart updates (#18889)! Thank you for your valuable addition to the project.

:rocket: Try it Out & Share Your Feedback!

We invite all of you to explore and leverage this latest release. Your feedback is instrumental in driving further advancements! If you encounter any issues, have questions, or simply want to share your experience, feel free to start a discussion here or open an issue.

Stay tuned for more exciting updates from the Ultralytics team and the fantastic YOLO community. Together, we continue pushing the boundaries of computer vision!

Release: v8.3.68

Happy coding,
The Ultralytics Team :heart: