New Release: Ultralytics v8.3.158

Discussion
, ,
1

:rocket: Ultralytics v8.3.158 Release: Smarter Image Classification & Streamlined Model Selection!

:glowing_star: Summary

We’re excited to announce Ultralytics v8.3.158! This update focuses on making image classification workflows more reliable and boosting the usability of our Streamlit inference tool. With cleaner preprocessing, smarter model selection, and improved tool stability, this release delivers a smoother and more accurate experience for everyone working with Ultralytics models.

Check out what’s new below and let us know your thoughts!

New Features & Improvements

:framed_picture: Image Classification Improvements

  • RGB-First Processing: Images for classification tasks are now expected in RGB format (not BGR), fixing color issues and leading to more accurate predictions.
  • Simplified Transforms: Outdated color channel reversal and legacy code have been removed. The image transform process is now more streamlined and easier to understand.
  • Cleaner Codebase: By eliminating legacy transforms, the code is easier to maintain and extend for future improvements.

PR by @Laughing-q:
Eliminate classification legacy transforms

:control_knobs: Streamlit Inference Tool Enhancements

  • Organized Model Selection: YOLO11 models are now sorted more intuitively by size and type, making the right choice quick and easy.
  • Smarter Selection List: Grayscale models are excluded, and custom models are now listed at the top for immediate access.
  • Improved Stability: Implemented a workaround to avoid possible PyTorch import errors, boosting reliability for all users.

PR by @RizwanMunawar:
Improve model list order in Streamlit solution

:hammer_and_wrench: Bug Fixes

  • PyTorch Import Issues: Addressed potential import errors in the Streamlit tool, making your inference experience smoother.

:bullseye: Purpose & Impact

  • Better Results: Accurate color handling directly leads to more reliable image classification outputs.
  • User-Friendly Tools: Streamlit users can now quickly find and select the best YOLO11 model for their needs.
  • Robust Workflow: Improvements in stability and codebase clarity mean less troubleshooting and more productivity for all.

:loudspeaker: Try It Out & Share Feedback

Update to Ultralytics v8.3.158 and let us know how it improves your experience! We invite everyone to try the updated classification features and Streamlit inference tool—your feedback directly shapes future releases.

As always, huge thanks to the YOLO community and all contributors who help make these improvements possible. The Ultralytics team is excited to see what you build next—happy experimenting!