ultralytics 8.3.131

v8.3.131 - `ultralytics 8.3.131` Optimize `grayscale` model inference pipeline (#20565)

on Python PyPI

🌟 Summary

This release brings enhanced support for grayscale images, improved handling of large and transparent images, a new YOLO11 C++ Triton Inference example, and important licensing and workflow updates. 🖤🖼️🚀

📊 Key Changes

• Grayscale Image Support:

  • Added a channels parameter to data loaders, enabling seamless inference and visualization with grayscale (single-channel) images.
  • Updated image reading and plotting utilities to handle grayscale images without errors.

• Large & Transparent Image Handling:

  • Removed the default image size limit in PIL, allowing safe processing of extremely large images.
  • Automatically converts images with transparency (e.g., PNG with alpha) to RGB to prevent saving errors.

• YOLO11 Triton C++ Example:

  • Introduced a new C++ example for deploying YOLO11 models on NVIDIA Triton Inference Server, including efficient FP16 inference and result visualization.

• Licensing & Workflow Improvements:

  • Added Ultralytics AGPL-3.0 license headers to Rust and C++ example files for legal clarity.
  • Updated GitHub workflow to prevent PRs labeled "TODO" from being auto-closed.

🎯 Purpose & Impact

• Broader Dataset Compatibility:

  • Users can now easily run inference on grayscale images, unlocking new applications in fields like medical imaging and scientific research.

• Robust Image Processing:

  • Eliminates errors when working with very large or transparent images, making the platform more reliable for diverse data sources.

• Production-Ready Deployment:

  • The new YOLO11 Triton C++ example helps developers deploy models efficiently in scalable, high-performance environments.

• Open-Source Clarity:

  • Clear licensing ensures users and contributors understand usage rights, while improved workflow keeps important contributions active.

This update makes Ultralytics tools more flexible, robust, and ready for advanced real-world use cases! 🚀

What's Changed

• Allow image compression >178.9M pixels by @RizwanMunawar in #20575
• Use ultralytics-actions>=0.0.73 by @glenn-jocher in #20581
• Add YOLO11 Triton CPP Example by @asdemirel in #20553
• Exempt TODO PRs from stale by @Y-T-G in #20571
• ultralytics 8.3.131 Optimize grayscale model inference pipeline by @Laughing-q in #20565

New Contributors

• @asdemirel made their first contribution in #20553

Full Changelog: v8.3.130...v8.3.131