New Release: Ultralytics v8.3.92

Discussion
,
1

:rocket: Announcing YOLO v8.3.92 Release!

We are excited to introduce the latest update to the Ultralytics YOLO ecosystem: YOLO v8.3.92! This release is packed with crucial improvements, new features, and refined documentation to enhance your overall experience. Whether you’re training for specific use cases, exporting models, or working on advanced visualizations, this version has something for everyone. :tada:

:glowing_star: Summary

The v8.3.92 release focuses on improving training flexibility, enhancing model export stability to TensorFlow edge environments, and refining documentation for more seamless navigation and usability. Additionally, new visual customization options are now available, further elevating your project’s visual coherence and adaptability.

:bar_chart: Key Changes

New Features

  • Customizable Detection Annotations:
    You can now customize text colors in detection result annotations using the new txt_color parameter. Deliver professional and project-specific visual outputs with ease! :artist_palette:
    PR: Expose txt_color parameter for Results plots by @zanaries

Improvements

  • Single-Class Training Fix:
    Resolved a cache error during single-class training by improving label processing logic, ensuring smoother workflows for users focused on specific object categories.
    PR: single_cls training cache error fix by @Y-T-G

  • TensorFlow Model Export Updates:
    Introduced ai-edge-litert>=1.2.0 as a dependency, improving the reliability and performance of TensorFlow model exports—especially for Edge AI applications.
    PR: Add ai-edge-litert>=1.2.0 to exporter.py by @glenn-jocher

  • Enhanced Jetson Compatibility:
    Fixed Python version check logic to prevent unnecessary dependency downgrades on Jetson devices, ensuring seamless usage.
    PR: Fix Autobackend Python version check by @Auc7us

Documentation Enhancements

:bullseye: Purpose & Impact

Each update in v8.3.92 was designed to improve your experience:

  1. Training Flexibility :glowing_star:
    Easily execute single-class training scenarios thanks to the cache error fix, empowering specialized object detection workflows.

  2. Effortless Exporting
    With upgraded TensorFlow model export capabilities, deploying across edge-AI environments is now more efficient and reliable.

  3. Broader Compatibility
    Jetson users benefit from a smoother experience, avoiding compatibility setbacks during setup.

  4. User-Friendly Documentation
    Better examples, formatting clarity, and fixed links ensure you can find guidance quickly without unnecessary friction.

  5. Customizable Visual Outputs :rainbow:
    Personalize text annotation colors for detection results, aligning them perfectly with your project’s branding or presentation needs.

:light_bulb: How to Get Started

:hammer_and_wrench: Contributions & Acknowledgments

We’re thrilled to welcome new contributors to the YOLO community:

  • @zanaries with their work on customizable annotation colors. :tada:
  • @Auc7us for improvements in the Jetson environment compatibility. :raising_hands:

Thank you to all contributors for making YOLO better with every release! Your efforts are invaluable. :folded_hands:

:speaking_head: We Value Your Feedback

We invite you to try out YOLO v8.3.92, explore the new features, and share your thoughts with us in the comments. If you encounter any issues or have suggestions, don’t hesitate to open a discussion or issue on GitHub. Together, let’s continue innovating and shaping the future of AI and computer vision! :rocket: