rapidocr

0 Fork 0

mirror of https://github.com/RapidAI/RapidOCR.git synced 2026-11-12 22:10:55 +08:00

Files

T

Jyotishman Das ae1d898295 fix(python): convert palette and CMYK images before OCR (#773 )

img_to_ndarray passed the raw PIL array straight to convert_img. For
palette (P) images that array holds palette indices, so a typical
two-colour PNG reaches the detector as an almost black image and no
text is found. CMYK images give four ink channels that were treated as
RGBA, so a CMYK JPEG also returned no result.

Convert P images to RGBA, which keeps any palette transparency for the
existing alpha handling, and CMYK images to RGB. Path, bytes and PIL
inputs all go through this function.

2026-11-12 21:57:38 +08:00

cal_rec_boxes

fix(word_box): score each word from its own characters (issue #749 ) (#750 )

2026-10-25 21:41:21 +08:00

ch_ppocr_cls

fix(cls): classify the middle of long lines with the v2.0 model (#748 )

2026-10-20 22:06:30 +08:00

ch_ppocr_det

perf(det): keep DetPreProcess mean/std in float32 (#756 )

2026-10-25 21:42:33 +08:00

ch_ppocr_rec

perf(torch): reduce rec logits on device before copying them to the CPU (#757 )