CATMuS Medieval Script Trainer

This project provides a training pipeline for models on the CATMuS Medieval dataset, which is a multilingual, multiscript medieval HTR dataset. The training portion comes largely from this notebook. It has been modified to align with CATMuS.

Installation

git clone https://github.com/wjbmattingly/catmus-train
cd catmus-train

# Install dependencies
pip install -r requirements.txt

Usage

python main.py --shuffle_seed 42 --select_range 1000 --batch_size 4 --epochs 10 --logging_steps 1000 --save_steps 1000 --save_limit 2 --compute_cer_flag --device mps:0 --scripts Caroline

List of Pre-trained Models

1. medieval-data/trocr-medieval-base
2. medieval-data/trocr-medieval-latin-caroline
3. medieval-data/trocr-medieval-castilian-hybrida
4. medieval-data/trocr-medieval-humanistica
5. medieval-data/trocr-medieval-textualis
6. medieval-data/trocr-medieval-cursiva
7. medieval-data/trocr-medieval-semitextualis
8. medieval-data/trocr-medieval-praegothica
9. medieval-data/trocr-medieval-semihybrida
10. medieval-data/trocr-medieval-print

Citation

If you use the CATMuS Medieval dataset, please cite the following paper:

BibTeX

@unpublished{clerice:hal-04453952,
  TITLE = {{CATMuS Medieval: A multilingual large-scale cross-century dataset in Latin script for handwritten text recognition and beyond}},
  AUTHOR = {Cl{\'e}rice, Thibault and Pinche, Ariane and Vlachou-Efstathiou, Malamatenia and Chagu{\'e}, Alix and Camps, Jean-Baptiste and Gille-Levenson, Matthias and Brisville-Fertin, Olivier and Fischer, Franz and Gervers, Michaels and Boutreux, Agn{\`e}s and Manton, Avery and Gabay, Simon and O'Connor, Patricia and Haverals, Wouter and Kestemont, Mike and Vandyck, Caroline and Kiessling, Benjamin},
  URL = {https://inria.hal.science/hal-04453952},
  NOTE = {working paper or preprint},
  YEAR = {2024},
  MONTH = Feb,
  KEYWORDS = {Historical sources ; medieval manuscripts ; Latin scripts ; benchmarking dataset ; multilingual ; handwritten text recognition},
  PDF = {https://inria.hal.science/hal-04453952/file/ICDAR24___CATMUS_Medieval-1.pdf},
  HAL_ID = {hal-04453952},
  HAL_VERSION = {v1},
}

APA

Thibault Clérice, Ariane Pinche, Malamatenia Vlachou-Efstathiou, Alix Chagué, Jean-Baptiste Camps, et al.. CATMuS Medieval: A multilingual large-scale cross-century dataset in Latin script for handwritten text recognition and beyond. 2024. ⟨hal-04453952⟩

License

This project is licensed under the MIT License.