ultralytics 8.3.166

v8.3.166 - `ultralytics 8.3.166` Standardize VisDrone autodownload structure (#21367)

on Python PyPI

🌟 Summary

This release focuses on improving dataset handling and clarity, with a major update to the VisDrone dataset structure and conversion process, along with several refinements to dataset configurations and evaluation logic. 🗂️✨

📊 Key Changes

• VisDrone Dataset Overhaul:
  • Standardized the folder structure for VisDrone downloads to use images/train, images/val, and images/test.
  • Enhanced the annotation conversion script to reliably generate YOLO-format labels, organized by data split.
  • Automated cleanup of original VisDrone folders after conversion for a tidier workspace.
• Consistent Dataset Configurations:
  • Standardized folder comments and formatting across all dataset YAML files for better readability and consistency.
  • Updated the listed download size for the tiger-pose dataset to reflect its actual size (now 49.8 MB).
• Evaluation Improvements:
  • Refined confusion matrix logic for oriented bounding box (OBB) tasks, ensuring low-confidence detections are excluded and thresholds are set appropriately.
• Documentation Enhancement:
  • Improved contributor information in the documentation for clearer author attribution.

🎯 Purpose & Impact

• 🧹 Cleaner Dataset Organization: The new VisDrone structure and automated cleanup make it easier for users to manage and use the dataset, reducing confusion and setup time.
• 📁 Consistency Across Datasets: Standardized comments and formatting help both new and experienced users understand dataset structures more quickly.
• 📏 More Accurate Evaluations: Improved handling of OBB confusion matrices leads to more reliable evaluation results, especially for advanced detection tasks.
• 📦 Transparent Downloads: Accurate dataset size listings prevent surprises during downloads and help users plan storage needs.
• 👥 Better Documentation: Enhanced author profiles in the docs foster community recognition and trust.

Overall, this update streamlines dataset usage, improves evaluation reliability, and enhances the user experience for both development and documentation. 🚀

What's Changed

• Fix tiger-pose.yaml download size by @Laughing-q in #21355
• Exclude low confidence detections in OBB confusion matrix by @Y-T-G in #21352
• Update mkdocs_github_authors.yaml by @glenn-jocher in #21366
• ultralytics 8.3.166 Standardize VisDrone autodownload structure by @glenn-jocher in #21367

Full Changelog: v8.3.165...v8.3.166