🌟 Summary

The Ultralytics v8.3.4 release introduces flexible updates to project dependencies and enhances several features for better usability and performance.

📊 Key Changes

• 🔗 Numpy Dependency Update: Removed the upper version limit on numpy, now supporting any version above 1.23.0.
• 🕒 CI Schedule Adjustment: Changed the daily build and test schedule to 08:00 UTC for better synchronization.
• 🚀 Improved Example Documentation: Added a detailed example for YOLOv8 using OpenVINO in C++.
• 🔢 Enhanced Error Handling in Export Formats: Introduced auto-correction mechanism for export format inputs.
• 🖥️ Device Selection Robustness: Improved parsing logic for device identifiers such as multiple GPUs.
• 📄 Documentation and Video Guides: Enhanced documentation with video tutorials on MobileSAM and SAM 2 models, and updated image links for better viewing.

🎯 Purpose & Impact

• Compatibility & Flexibility: Unpinning the numpy version boosts compatibility with newer versions, simplifying dependency management and reducing potential conflicts with other packages.
• Optimized Workflow: The change in CI schedule aligns better with daily resource availability and developer workflows, enhancing continuous integration efficiency.
• User Experience: Improved error handling during export ensures users receive guidance on available formats, minimizing confusion and errors.
• Accessibility & Education: Video tutorials and clearer documentation aid both new and experienced users in understanding and leveraging advanced model functionalities, fostering a wider user base.
• Performance Optimization: The example using OpenVINO can potentially enhance inference speed and support for various hardware profiles, such as Intel platforms.

What's Changed

• Fix Raspberry Pi CI by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16620
• Update scheduled CI to 08:00 UTC by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16622
• Add https://youtu.be/yXQPLMrNX2s to docs by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16626
• Fix overridden train_args when close_mosaic by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/16627
• Add contributor name for example: YOLOv8 OpenVINO CPP Inference by @rlggyp in https://github.com/ultralytics/ultralytics/pull/16632
• Auto-correct for export formats by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16625
• Empty index [0,,1] robust device selection by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16631
• Improve scikit-learn support for exports by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16638
• Fixing predict docs by @ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/16637
• SAM: Fix labels not being used in predict mode. by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/16642
• Minor fix in docs for python admonition and code blocks by @jk4e in https://github.com/ultralytics/ultralytics/pull/16646
• Update image URLs in mobile-sam.md and sam-2.md. by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16654
• Replace YOLO11-seg table with macro by @jk4e in https://github.com/ultralytics/ultralytics/pull/16647
• ultralytics 8.3.4 unpin numpy<2.0.0 by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16655

Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.3...v8.3.4