🌟 Summary
This release streamlines image classification workflows and enhances the model selection experience in the Streamlit inference tool, making both more reliable and user-friendly. 🖼️🚀
📊 Key Changes
- Image Classification Improvements:
- Updated image preprocessing to expect images in RGB format (not BGR), ensuring accurate color handling.
- Removed outdated color channel reversal and legacy code from the classification pipeline.
- Simplified and clarified how image transforms are applied for classification tasks.
- Streamlit Inference Tool Enhancements:
- Improved the model selection dropdown: YOLO11 models are now sorted more intuitively by size and type.
- Grayscale models are excluded from the selection list for clarity.
- Custom models now appear at the top of the dropdown for easier access.
- Added a workaround to prevent potential PyTorch import errors, increasing tool stability.
🎯 Purpose & Impact
- 🖼️ Better Classification Results: Ensures images are processed in the correct color format, reducing errors and improving reliability for users running classification tasks with YOLO models.
- 🧹 Cleaner Codebase: Removes legacy code, making the system easier to maintain and extend for developers.
- 👩💻 Improved Usability: Streamlit users can now find and select the right YOLO11 model more easily, with a cleaner and more organized interface.
- 🛡️ Increased Stability: Fixes potential import issues with PyTorch, reducing crashes and making the inference tool more robust.
Overall, this update delivers a smoother, more accurate, and user-friendly experience for both developers and end users working with Ultralytics models and tools.
What's Changed
- Improve model list order in
Streamlitsolution by @RizwanMunawar in #21058 ultralytics 8.3.158Eliminate classification legacy transforms by @Laughing-q in #21089
Full Changelog: v8.3.157...v8.3.158