---
library_name: PyLaia
license: mit
tags:
- PyLaia
- PyTorch
- atr
- htr
- ocr
- historical
- handwritten
metrics:
- CER
- WER
language:
- fr
- la
datasets:
- Teklia/Himanis
pipeline_tag: image-to-text
---

# PyLaia - Himanis

This model performs Handwritten Text Recognition in French on medieval documents.

## Model description

The model was trained using the PyLaia library on two medieval datasets:
* [Himanis](https://demo.arkindex.org/browse/5000e248-a624-4df1-8679-1b34679817ef?top_level=true&folder=true) (French)
* [HOME Alcar](https://demo.arkindex.org/browse/46b9b1f4-baeb-4342-a501-e2f15472a276?top_level=true&folder=true) (Latin)

For training, text-lines were resized with a fixed height of 128 pixels, keeping the original aspect ratio.

An external 6-gram character language model can be used to improve recognition. The language model is trained on the text from the Himanis training set.

## Evaluation results

On Himanis text lines, the model achieves the following results:

| set   | Language model | CER (%)    | WER (%) | N lines   |
|:------|:---------------| ----------:| -------:|----------:|
| test  | no             | 9.87       |   29.25 |     2,241 |
| test  | yes            | 8.87       |   24.37 |     2,241 |

## How to use?

Please refer to the [documentation](https://atr.pages.teklia.com/pylaia/).

## Cite us!

```bibtex
@inproceedings{pylaia-lib,
    author = "Tarride, Solène and Schneider, Yoann and Generali, Marie and Boillet, Melodie and Abadie, Bastien and Kermorvant, Christopher",
    title = "Improving Automatic Text Recognition with Language Models in the PyLaia Open-Source Library",
    booktitle = "Submitted at ICDAR2024",
    year = "2024"
}
```