🌟 Summary
The v8.3.23 release of Ultralytics YOLO introduces stability improvements by fixing a bug in the data conversion process, updating versioning, and refining user settings and documentation.
📊 Key Changes
- Version Update: The software version is updated from 8.3.22 to 8.3.23.
- Bug Fix in Data Conversion: Ensures
yolo_bbox2segmentskips generating empty segment lists, preventing potential errors. - Reduced Python Warnings: Edited Python version check to minimize console spam.
- Documentation Update: Corrected export format examples for INT8 quantization, aligning with TensorRT capabilities.
- W&B Logger Default: Weights & Biases logging is now disabled by default.
- Environment Detection: Enhanced accuracy for identifying Jupyter environments.
🎯 Purpose & Impact
- 🚀 Improved Stability: Fixing the bug in data conversion leads to more reliable performance during bounding box to segment transformations.
- 📉 Cleaner Console: Less console clutter from Python checks makes for a smoother user experience.
- 📝 Clearer Documentation: Updated docs guide users on proper export procedures, easing model deployment tasks.
- 📈 Optimized Resource Use: Disabling Weights & Biases by default reduces unnecessary compute and network usage unless needed.
- 🧠 Reliable Environment Behavior: Accurate environment detection prevents misidentification in diverse setups, adapting better to where the software runs.
What's Changed
- Fix Python warning spam by @Y-T-G in #17162
- Fix inaccurate example in Export docs by @Y-T-G in #17161
- Default W&B setting
Falseby @glenn-jocher in #17164 ultralytics 8.3.23fixbbox2segmentwhen no segments generated by @Laughing-q in #17157
Full Changelog: v8.3.22...v8.3.23