v514
v514
Removed
November 9, 2026
Improved PDF redaction and style detection performance.
Graph_dag: map an EXIF-rotated JPEG's OCR boxes to its stored pixels.
V5 PDF synthesis is now the default - New datasets now use V5 PDF synthesis, which matches the font, size, and colors of the original text. V5 is no longer labeled experimental. V1 is still available as a Legacy option, and existing datasets keep their current setting.
Fixed PDF text reflow to preserve redaction completeness, text order, layer visibility, and surrounding content.
