BDRC/monlamai-handwritten

Tibetan OCR — MonlamAI handwritten (synthetic pages)

11,755 synthetic page images of handwritten Tibetan dbu-med (u-med) text with line-accurate Unicode transcriptions (line breaks preserved). Each page is reconstructed from single-line handwritten crops produced during the MonlamAI project (see below) composed in natural reading order into multi-line pages (~6 rows per page) with margins and inter-line gaps. Because the layout is synthetic, this set is intended for training only, not validation/evaluation.

The original dataset was created by printing forms to be filled by scribes with a sentence to transcribe in a delimited area, and then scan this area and associate it with the corresponding sentence.

Part of the BDRC Tibetan OCR release; see the model BDRC/tibetan-ocr and the companion real-transcription set BDRC/monlamai-transcriptions.

At a glance

Images (pages)	11,755 (synthetic, reconstructed)
Total size	2.05 GiB
Script	u-med (handwritten)
Technology	manuscript / handwritten
Line breaks preserved	Yes
Source of transcriptions	MonlamAI line ground truth (MonlamAI_LTT)
Intended use	training only (synthetic layout)
License	CC0-1.0

How the pages are made

Line crops are composed into pages by scripts/build_monlam_synth.py (bec-ocr-training):

• Background is sampled from the crops (average of the non-black pixels) so the page matches the source paper tone; clean/bilevel scans snap to white.
• Segmentation-mask removal: the pure-black padding the line-segmentation model leaves around a crop is replaced with the background; it is skipped for bilevel crops (whose text is pure black) so glyphs are never touched.
• New W / IE / VE / I ids carry the source id plus a _SYNTH02 suffix, under the same MW as the source.

Caveats

• Synthetic layout, not real page geometry — use for training, not eval.

Contents & schema

One row per page. Columns:

Column	Description
image	embedded synthetic page (HF Image)
transcription	Tibetan Unicode ground truth, line breaks preserved
id	unique row id
mw_id	BDRC version id of the source volume
technology	manuscript
script	ume
script_4	writing-style family (gyuyig/tsugdri)

from datasets import load_dataset
ds = load_dataset("BDRC/monlamai-handwritten", split="train")
print(ds[0]["transcription"]); ds[0]["image"]

License, attribution & image rights

• License: CC0-1.0 — use, modify, redistribute freely.
• Attribution requested (not required): please cite the Buddhist Digital Resource Center (BDRC) and MonlamAI.

Citation

@misc{bdrc_monlamai_handwritten_2026,
  title  = {Tibetan OCR --- MonlamAI handwritten (synthetic pages)},
  author = {Buddhist Digital Resource Center and MonlamAI},
  year   = {2026},
  howpublished = {Hugging Face},
  note   = {https://huggingface.co/datasets/BDRC/monlamai-handwritten}
}