roux.elie · Baselines · Published August 30, 2026
TiBLINE : Tibetan pecha layout detection
Layout Analysis
Description
TiBLINE is a layout-analysis model for Tibetan pecha (traditional loose-leaf) documents. It performs text-line detection — baselines and the surrounding line regions — on folios written in the Tibetan script, covering both woodblock prints (xylographs) and manuscripts. It is designed as the layout/segmentation stage of an OCR pipeline: it locates the individual lines on each page so a downstream text-recognition model can transcribe them.
The model was trained on 1,085 annotated pages (plus 106 validation pages, ~8,300 lines) of BDRC material spanning a range of pecha layouts, page dimensions and image qualities, including bound volumes, modern reprints and historical blockprints.
Best suited for: Tibetan pecha and pecha-style pages with one or several horizontal lines per folio. Limitations: this is a line/layout detector only (it does not recognise text), and accuracy may drop on heavily degraded scans, cursive (Ume) hands, dense interlinear annotations, or non-pecha page formats.
Created by BDRC (Buddhist Digital Resource Center) in collaboration with Dharmaduta, as part of the project "The BDRC Etext Corpus", funded by the Khyentse Foundation.
Open in Transkribus
Low loss5.14% loss
Loss indicates how far the predicted text regions deviate from the ground truth (lower is better). This model achieved 5.14% on its validation set. A loss below 10% generally indicates reliable baseline detection. Trained on a broad range of page layouts, this model should generalise well. Complex or unusual structures may still require fine-tuning.
Layout detection quality depends heavily on your document's structure. Pages with columns, marginalia, or non-standard layouts may produce different results.
Lines8,328
Training Pages1,085
Model ID627409