ultralytics 8.3.139

v8.3.139 - `ultralytics 8.3.139` New `DataExportMixin` class for Metrics and Results exports (#20546)

on Python PyPI

🌟 Summary

This release introduces a powerful new way to export YOLO validation metrics and prediction results in multiple formats, making it much easier to analyze, share, and integrate your results. 📊✨

📊 Key Changes

• Unified Data Export: Added a new DataExportMixin class that lets you export metrics and results as DataFrame, CSV, XML, HTML, JSON, or directly to an SQLite database.
• Broader Export Support: Integrated these export features into all major metrics/results classes, including detection, segmentation, classification, pose, and oriented bounding box tasks.
• Improved Documentation: Updated guides to show how to use the new export options and clarified available formats.
• Better Testing: Enhanced test coverage to ensure all export methods work reliably.
• Faster Dependency Management: Now uses the "uv" installer for Python packages (when available), making setup faster and smoother.
• Raspberry Pi Test Optimization: Skips slow similarity search tests on Raspberry Pi for a more reliable experience.
• TFJS Export Fix: Automatically disables group convolutions when exporting models to TensorFlow.js, preventing errors for web deployments.

🎯 Purpose & Impact

• Simplifies Data Handling: Effortlessly export your validation results and metrics in the format that best fits your workflow—no more manual conversions or workarounds.
• Boosts Productivity: Makes it easier to analyze results in tools like Excel, databases, or web dashboards, and to share findings with collaborators.
• Speeds Up Setup: Faster package installations mean less waiting and more doing, especially for users with "uv" installed.
• Enhances Compatibility: Smoother model exports for TensorFlow.js and better support for Raspberry Pi users.
• Future-Proof: Lays the groundwork for more flexible integrations and reporting in Ultralytics workflows.

Overall, this update is a big step forward for usability, efficiency, and integration in the Ultralytics ecosystem! 🚀

What's Changed

• Use uv in check_requirements by @glenn-jocher in #18116
• Disable similarity search tests on RaspberryPI by @RizwanMunawar in #20690
• Fix YOLO11 TFJS inference error by @Y-T-G in #20687
• ultralytics 8.3.139 New DataExportMixin class for Metrics and Results exports by @RizwanMunawar in #20546

Full Changelog: v8.3.138...v8.3.139