New Release: Ultralytics v8.3.47

:rocket: New Ultralytics Release: v8.3.47 :tada:

We are thrilled to announce the release of Ultralytics v8.3.47! This latest version comes packed with exciting improvements, particularly focused on leveling up YOLO’s classification capabilities. If you’ve been working on classification tasks, you’ll find new features that enhance both development and deployment ease. :bulb:


:star2: Summary

The v8.3.47 release brings:

  • A more versatile YOLO Classification Module with enhanced export capabilities and improved loss handling.
  • Streamlined documentation updates, making feature implementation and workflows easier for users.
  • Fixes and optimizations to ensure your development experience continues to be frictionless and efficient.

:key: Key Changes

Enhanced YOLO Classification Module

  • :hammer_and_wrench: Export functionality added: You can now enable exports for classification models using export=True.
  • :arrows_counterclockwise: Improved training/inference workflow: Classification head now smoothly switches between training and inference, with built-in softmax during inference.
  • :bar_chart: Loss function enhancements: The classification loss computation has been updated for consistency when processing tuple-based predictions, ensuring greater accuracy.
  • :gear: Streamlined predictions handling: Enhanced handling of tuple-based outputs in post-processing, increasing robustness during model inference.

Documentation Updates

  • :writing_hand: Clarified the functionality of the copy_paste augmentation: clearly indicating it applies only to segmentation labels.
  • :hammer_and_wrench: Tidied up YAML examples and eliminated redundant documentation entries for workflows that are easier to follow.
  • :books: Better typo management and content presentation, allowing developers to save time when navigating the docs.

:dart: Purpose & Impact

For End Users

  • :white_check_mark: Easier Model Deployment: With streamlined export workflows, deploying YOLO classification models into production becomes seamless.
  • :mag: More Robust Classification: Real-world classification performance benefits from smoother training logic and inference handling.

For Developers

  • :wrench: Developer-first Improvements: Updated YAML + typo fixes eliminate tricky roadblocks during implementation, boosting efficiency.
  • :sparkles: Simplified Augmentation Rules: Refined explanations and examples make advanced features like copy_paste easier to use correctly.

Bottom line: This version ensures better versatility and reliability for classification use cases while strengthening supporting documentation and clarity.


:arrows_counterclockwise: What’s Changed?

Here’s a detailed breakdown of the updates in this release:

Full Changelog: Compare v8.3.44…v8.3.47


:tada: Try It Out & Share Your Feedback!

We encourage you to upgrade to v8.3.47 today and explore the new classification features along with the improved documentation. As always, your feedback is invaluable—please share your thoughts, experiences, or suggestions in the Ultralytics Discussions Board. :speech_balloon:

Ready to get started? Download v8.3.47 here. :rocket:

We can’t wait to see how this release helps you push your projects to the next level. Thank YOU for being a valuable part of the Ultralytics community! :star2: