ultralytics/ultralytics v8.4.206 on GitHub

🌟 Summary

Ultralytics 8.4.206 improves PaddleOCR’s handling of upside-down text and makes OCR/rotated-box labels easier to read. 🔄

Reads upside-down text with PaddleOCR: The OCR solution now uses a text-line orientation classifier to identify text that may be upside down. It tries those crops in both orientations and keeps the reading with higher confidence. This also helps with bottom-to-top text when its crop is oriented upside down.
Updates OCR dependencies: PaddleOCR now requires paddleocr>=3.7 and PaddlePaddle 3.1 or later. Custom recognizers are unchanged and do not get automatic orientation correction.
Improves label placement: OCR and oriented bounding-box labels now sit above the top-left of the box enclosing the polygon, rather than covering rotated text or appearing at an arbitrary corner. OCR labels also use black text on green for contrast.
Refreshes CI runner labels: GitHub Actions now uses macos-latest and ubuntu-26.04-arm labels, reducing future maintenance as runner images change.
Clarifies NXP documentation: The guide describes NXP’s Neutron SDK source as a pip package index instead of linking to a page that returns an error in browsers. Export behavior is unchanged.

More reliable OCR: PaddleOCR users should get better results on text that appears upside down, with the more confident reading selected automatically.
Clearer visualizations: Labels are less likely to obscure text or polygons, making plotted results easier to inspect.
No change for custom OCR backends: Users supplying their own recognizer remain responsible for handling text orientation.
No change to NXP exports: The documentation link was corrected; pip-based SDK installation continues to work.

Use macos-latest and ubuntu-26.04-arm runner labels by @glenn-jocher in #27080
Place OCR and OBB labels above the box instead of over rotated text by @glenn-jocher in #27081
Drop the NXP package index link that only answers pip by @glenn-jocher in #27082
ultralytics 8.4.206 Read upside-down and bottom-to-top text in OCR by @glenn-jocher in #27083

Full Changelog: v8.4.205...v8.4.206