BAYBARS — Line-Level HTR Ground-Truth for Arabic historical manuscripts (Oriental script)

Dataset Description

BAYBARS is a specialized dataset for Handwritten Text Recognition (HTR) focusing on Arabic historical manuscripts. It contains a curated selection of pages from the Sīrat Baybars (the Epic of Baibars) transcribed in the Oriental script (Levantine Middle Arabic).

This HuggingFace dataset provides cropped line-level images paired with their transcriptions and rich metadata for 3 Arabic notebooks (with various hands). It is designed as a ready-to-use resource for Arabic historical HTR.

Notebooks	3
Pages	546
Lines	15,567
Words	100,502
Characters	502,534

The full page-level dataset (PageXML + full-page images) is available on GitLab.

Source Documents

Three notebooks of the Sīrat Baybars in the Damascene recension, each containing between 163–200 booklets:

• baybars-nt-ms1
• baybars-nt-ms2
• baybars-nt-ms3

Original images are available on Gallica (BnF).

Usage

from datasets import load_dataset

# Load the full dataset
ds = load_dataset("calfa-ai/baybars")

# Access a sample
sample = ds["train"][0]
sample["image"].show()
print(sample["transcription"])

# Filter by manuscript
ms1 = ds["train"].filter(lambda x: x["manuscript_name"].startswith("baybars-nt-ms1"))

Annotation Process

Annotations were made on the Calfa Vision platform as part of a hackathon (September 2023 – July 2024), involving sixteen participants from five Lebanese universities, one French institution, and the French Institute of the Near East (IFPO). Calfa developed the HTR model and performed automatic transcription of the entire corpus, achieving a recognition accuracy of 96.5–99%.

Transcription Guidelines (Summary)

The transcription guidelines follow the RASAM specification: transcriptions preserve the text as found in the image, including variant spellings, while excluding vocalization marks unless structurally significant. See RASAM 1 and original paper for details.

License

This dataset is released under the Etalab Open Licence 2.0.

Acknowledgements

The Baybars Hackathon was organized as part of the ANR program Popular Literatures of the Levant (LiPoL). It was made possible thanks to the support of BnF and Ifpo, as well as the Université Saint-Joseph, which hosted us at the Institut des Lettres orientales/ILO. The training session and the production of the HTR model were conducted by Calfa.