RaphaelBfr/morphology4metrology-bnf2813
Dataset for BnF, fr. 2813 — Grandes Chroniques de France Line-level dataset used in the experiments for the paper Leveraging Morphology for Historical Script Metrological Analysis (ICDAR 2026). Annotator: Malamatenia Vlachou Efstathiou Description This dataset contains polygonal line extractions (with alpha transparency) and line-level transcriptions from the codex Paris, BnF, fr. 2813 (Grandes Chroniques de France). The identifier btv1b84472995 refers to the ark… See the full description on the dataset page: https://huggingface.co/datasets/RaphaelBfr/morphology4metrology-bnf2813.
Dataset for BnF, fr. 2813 — Grandes Chroniques de France
Line-level dataset used in the experiments for the paper [Leveraging Morphology for Historical Script Metrological Analysis](https://arxiv.org/abs/2606.09446) (ICDAR 2026).
Annotator: Malamatenia Vlachou Efstathiou
Description
This dataset contains polygonal line extractions (with alpha transparency) and line-level transcriptions from the codex [Paris, BnF, fr. 2813](https://gallica.bnf.fr/ark:/12148/btv1b84472995) (Grandes Chroniques de France).
The identifier btv1b84472995 refers to the ark ID of this manuscript in Gallica.
This Hugging Face repository mirrors the dataset.zip folder from the official Zenodo release (v2).
For the full Zenodo package (including HTR ground truth with page images and ALTO XML), see the Zenodo record.
Folder structure
btv1b84472995/
├── annotation.json
├── images/
│ ├── <page_folder>/
│ │ ├── <line_id>.png
│ │ └── ...
│ └── ...
├── charset.json # generated by the training code if absent
├── charset_accent.json
└── charset_without_accent.jsonannotation.json
JSON object keyed by line image filename:
{
"<image_id>": {
"split": "train",
"label": "A beautiful calico cat.",
"line": "DefaultLine",
"zone": "MainZone#1",
"script": "RaouletOrleans",
"folio": "1r",
"gp": "GP1",
"doc": "HT1"
}
}Usage with morphology4metrology
- Download this dataset:
pip install huggingface_hub
hf download RaphaelBfr/morphology4metrology-bnf2813 --repo-type dataset --local-dir /path/to/datasets/btv1b84472995- Set
datasets_pathin `datasets/config.json`:
{
"datasets_path": "/path/to/datasets"
}- Use
--data_folder btv1b84472995in the training scripts.
Pretrained weights: RaphaelBfr/morphology4metrology
Training scripts: `scripts/btv1b84472995/`
What's in Zenodo v2 (not all included here)
The Zenodo record also provides:
- `btv1b84472995_GT.zip` — HTR ground truth with full-page Gallica images and ALTO XML annotations (eScriptorium)
- `btv1b84472995_metadata.csv` — metadata for newly annotated folios (v2 adds 95 folios)
- Two versions of
annotation.jsonon Zenodo (with / without hyphenation for word breaks at end of lines)
Image credit
High-resolution source images: Source [gallica.bnf.fr](https://gallica.bnf.fr) / Bibliothèque nationale de France
License
Creative Commons Attribution 4.0 International (CC-BY-4.0)
Citation
If you use this dataset, please cite the Zenodo record:
@dataset{vlachou2026bnf2813,
author = {Vlachou Efstathiou, Malamatenia},
title = {Dataset for BnF, fr. 2813 - Grandes Chroniques de France},
year = {2026},
publisher = {Zenodo},
doi = {10.5281/zenodo.18745702},
url = {https://doi.org/10.5281/zenodo.18745702}
}And the paper:
@inproceedings{vlachou2026metrology,
title = {Leveraging Morphology for Historical Script Metrological Analysis},
author = {Vlachou-Efstathiou, Malamatenia and Baena, Raphael and
Stutzmann, Dominique and Aubry, Mathieu},
booktitle = {Document Analysis and Recognition -- ICDAR 2026},
publisher = {Springer},
year = {2026}
}