Tasks/OCR & Document Understanding

OCR & Document Understanding

লিখন শনাক্তকরণ ও নথি বিশ্লেষণ

Recognising printed and handwritten Bangla text at word, line, document and scene level, and extracting information from document images. Isolated character and digit classification is out of scope — see TODO-data.md.

1 dataset·45 papers
Source paperLink
BanglaWriting
Offline Bangla handwriting pages with manually annotated word boxes and Unicode text. The dataset license is CC BY-NC 3.0; the related 2021 Data in Brief article has a separate CC BY 4.0 license.
260 pages, 21,234 annotated words, 32,787 charactersCC BY-NC 3.02020Mendeley Data v1 (2020-10-19)
✓ verified 2026-10-23