---
license: mit
language:
- lat
- fr
- it
- spa
- ca
datasets:
- CATMuS/medieval
tags:
- trocr
- image-to-text
widget:
- src: >-
    https://huggingface.co/medieval-data/trocr-medieval-cursiva/resolve/main/images/cursiva-1.png
  example_title: Cursiva 1
- src: >-
    https://huggingface.co/medieval-data/trocr-medieval-cursiva/resolve/main/images/cursiva-2.png
  example_title: Cursiva 2
- src: >-
    https://huggingface.co/medieval-data/trocr-medieval-cursiva/resolve/main/images/cursiva-3.png
  example_title: Cursiva 3
---

![logo](logo-cursiva.png)

# About

This is a TrOCR model for medieval Cursiva. The base model was [microsoft/trocr-base-handwritten](https://huggingface.co/microsoft/trocr-base-handwritten). The model was then finetuned to Caroline: [medieval-data/trocr-medieval-latin-caroline](https://huggingface.co/medieval-data/trocr-medieval-latin-caroline). From a saved checkpoint, the model was further finetuned to Cursiva.

The dataset used for training was [CATMuS](https://huggingface.co/datasets/CATMuS/medieval).

The model has not been formally tested. Preliminary examination indicates that further finetuning is needed.

Finetuning was done with finetune.py found in this repository.

# Usage

```python
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image
import requests

# load image from the IAM database
https://huggingface.co/medieval-data/trocr-medieval-cursiva/resolve/main/images/cursiva-1.png
image = Image.open(requests.get(url, stream=True).raw).convert("RGB")

processor = TrOCRProcessor.from_pretrained('medieval-data/trocr-medieval-cursiva')
model = VisionEncoderDecoderModel.from_pretrained('medieval-data/trocr-medieval-cursiva')
pixel_values = processor(images=image, return_tensors="pt").pixel_values

generated_ids = model.generate(pixel_values)
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
```

# BibTeX entry and citation info

## TrOCR Paper

```tex
@misc{li2021trocr,
      title={TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models}, 
      author={Minghao Li and Tengchao Lv and Lei Cui and Yijuan Lu and Dinei Florencio and Cha Zhang and Zhoujun Li and Furu Wei},
      year={2021},
      eprint={2109.10282},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
```

## CATMuS Paper

```tex
@unpublished{clerice:hal-04453952,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},
}
```