New Release: Ultralytics v8.3.168

Discussion
, ,
1

:rocket: Ultralytics v8.3.168 Released — Unified Prediction Export, Adaptive Annotations, and More!

:glowing_star: Summary

Ultralytics v8.3.168 is here! This release delivers a major upgrade to how YOLO models handle and export predictions, brings in a streamlined approach for image annotation, and packs enhancements into our documentation and user-facing tools for a smoother, more consistent experience across the platform. Whether you’re a developer, researcher, or just getting started with computer vision, there’s something for everyone in this update.

:bar_chart: Key Changes

New Features

  • Unified Prediction Export:
    All YOLO models (including detection, segmentation, pose, OBB, and RT-DETR) now use a refactored, robust export system for COCO JSON. This ensures consistent handling of image metadata and scaling, simplifying downstream evaluation and metric reporting.

  • Adaptive Annotation Labels:
    The new adaptive_label method replaces circle_label and text_label, letting you draw flexible, context-aware labels (circle or rectangle) in one line—making annotation easier and reducing code clutter.

  • Streamlit Image Support:
    The Ultralytics YOLO Streamlit app now allows image uploads for real-time inference, expanding accessibility so you can quickly test models on images, video, or live webcam feeds—all from your browser.

Improvements

  • Documentation Updates:

  • Scoped Imports in VisualAISearch:
    Streamlined imports in VisualAISearch solutions for faster and more reliable performance.

  • Professionalism & Accuracy:
    Corrected author information in YOLO11 documentation so contributors receive proper credit.

:bullseye: Purpose & Impact

  • Consistency & Reliability:
    All YOLO models now export results in a consolidated, standardized way—improving usability and robustness for metric reporting and pipelines.
  • Smoother Annotation Workflow:
    The unified adaptive_label simplifies the toolchain for users and developers, supporting more flexible annotations with less code.
  • Enhanced Accessibility:
    With image upload in the Streamlit app, hands-on testing works right out of the box—ideal for demos and rapid prototyping.
  • Cleaner Developer Experience:
    Documentation is more actionable and transparent, while codebase improvements make it easier for all users to integrate and extend.
  • Improved Citation:
    Ensuring all contributors and sources are credited maintains trust and professionalism across the Ultralytics community.

:wrench: What’s Changed

A quick summary of notable PRs included in v8.3.168:

Full changelog:
Compare v8.3.167…v8.3.168 on GitHub

Release Page:
Ultralytics v8.3.168 Release Notes

:vertical_traffic_light: Get Involved!

We invite the community to update to v8.3.168, explore the new features—especially the unified export and adaptive annotation labels—and let us know how they improve your workflow. Please share any feedback, suggestions, or questions either in this topic or directly in the repository’s Discussions section.

As always, a huge thank you goes to every contributor, tester, and user who helps make Ultralytics and YOLO better each day. This progress is a testament to the amazing efforts of the whole YOLO community and the Ultralytics team!

Happy experimenting! :tada: