Skip to content

Indra OCR 1.1.2

Latest
Latest

Choose a tag to compare

View all tags
@VikParuchuri VikParuchuri released this 20 Oct 01:40
· 3 commits to main since this release
v1.1.2
0e96b93

Install: pip install -U indra-ocr. Same models as 1.1.0 (Hugging Face datalab-to/indra-ocr, tag v1.1.0).

  • PDFs with legacy-font text layers. Many Hindi PDFs use legacy fonts (Kruti Dev, DevLys), whose text layer stores Latin codes in place of Devanagari, so Indra read them as Latin garbage. Indra now checks the script of a few text-layer lines against the line images; when the layer is Latin while the page shows another script, OCR reads the page. On such Hindi pages, word accuracy goes from 0.01-0.20 to 0.85-0.89. It adds about 5 ms per text-layer page.
  • If a PDF's text layer still reads wrong, --mode ocr reads the page itself.
Assets 2
Loading