Ultralytics YOLO

🌟 Summary

The Ultralytics v8.3.22 release brings significant improvements by integrating the SAM 2.1 model, enhancing segmentation capabilities, and implementing various fixes and updates for better user experience and functionality.

📊 Key Changes

• SAM 2.1 Integration: Introduces support for the SAM 2.1 model across all scales, upgrading pre-trained weights to version 8.3.0.
• Device Handling Fix: Improves logic for exporting models to TensorRT, ensuring correct device processing.
• Configuration Updates: Streamlines solution-specific default configurations directly within the code.
• Binder Integration: Adds a Binder badge to allow running Ultralytics in an interactive Jupyter notebook environment.

🎯 Purpose & Impact

• Improved Segmentation: SAM 2.1 enhances segmentation accuracy with advanced algorithms and features like spatial memory handling and temporal encoding, benefiting users needing precise object segmentation. 🎨
• Robust Exporting: Fixes in device handling bolster reliability when exporting models, preventing potential errors and aiding in smoother operation. ⚙️
• User Experience Enhancement: The consolidation of configuration management reduces complexity, providing a more seamless user setup process. 🛠️
• Accessibility: The Binder integration makes Ultralytics more accessible, allowing users to easily experiment with the software in a flexible, online environment. 🌐

What's Changed

• Fix DLA export when device=None by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/17128
• Enable default cfg for similar args in solutions. by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/17112
• Add Binder Notebook badge by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/17074
• ultralytics 8.3.22 SAM2.1 integration by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/17131

Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.21...v8.3.22