New Release: Ultralytics v8.3.37

Discussion
,
1

:rocket: Ultralytics v8.3.37 Release Announcement

We’re thrilled to announce the release of Ultralytics v8.3.37! :tada: This update brings exciting new features, key improvements, and important fixes designed to enhance your experience and streamline workflows. Here’s everything you need to know about this release:

:star2: Summary

The new version, v8.3.37, focuses on improving export functionality, refining model operation modes, and enhancing the user experience with updates to documentation and annotation utility resources.

From better handling of TensorRT exports to the introduction of an eval method for models, this release ensures that both the new and seasoned users of Ultralytics benefit from optimized operations and simpler workflows.

:bar_chart: Key Changes

1. TensorRT Auto-Workspace Size

  • Exporting models with TensorRT just got easier! The workspace parameter is now set to None by default, enabling auto-managed workspace allocation, saving both time and reducing configuration complexities.
  • PR: #17748 by @Burhan-Q

2. Label Padding Adjustment

  • Fixed the label augmentation to properly calculate vertical and horizontal padding when center=False.
  • This improvement ensures higher annotation accuracy, critical for training and evaluating models effectively.
  • PR: #17728 by @Y-T-G

3. Model Evaluation Mode

  • Added a method, model.eval(), to easily switch models between training and evaluation modes. This helps ensure consistent performance during evaluations and simplifies performance metric generation.
  • PR: #17754 by @Laughing-q

4. Documentation Updates

  • New video tutorials were added in the documentation for hand keypoint estimation (Video 1) and annotation utilities (Video 2).
  • Standardized dataset configuration references (e.g., updates like replacing obsolete coco-seg.yaml with coco.yaml) ensure clearer annotation and setup steps.
  • PRs:

5. Dependency Updates

  • Effective maintenance and improvements applied across GitHub workflows, like moving to more robust dependencies (e.g., bumping astral-sh/setup-uv version).
  • PR: #17753 by @dependabot[bot]

:dart: Purpose and Impact

  • Ease of Exporting Models: The TensorRT auto-workspace size adds seamlessness to model exporting workflows.
  • Enhanced Annotation Accuracy: Optimized label padding contributes to better training results.
  • Ease of Model Assessment: The ability to toggle into evaluation mode benefits users dealing with rigorous model testing and validation.
  • Better Documentation: Updated guides and robust video tutorials enhance user understanding, encouraging correct feature adoption.

:hammer_and_wrench: Full Release Details

Here’s the complete list of changes:

Changelog: v8.3.36…v8.3.37
Release URL: Ultralytics v8.3.37

:loudspeaker: We Need Your Feedback!

We encourage you to try out this new release and explore its features! If you spot any issues, have suggestions, or simply want to share your experiences, feel free to join the conversation on our GitHub Discussions or open a new issue.

Thank you for being a part of the YOLO community and continually helping us improve. See you in the next update! :rocket:

Happy exploring!
The Ultralytics Team