README.md

---
pretty_name: "GenHisDoc"
repos :
  - repo : https://github.com/Anarchiviste/GenHisDoc_Lab/tree/main

language:
  - "en"
  - "fr"

tags:
  - "computer vision"
  - "historical documents"
  - "detection"
  - "python"

license: "cc-by-nc-sa-4.0"

size_categories: "n<10k"

annotations_creators:
  - "found"
  - "corrected"
  - "annotated"
  - "Jules Musquin"
---

# GenHisDoc - Generalistic Historical Document Dataset - 9448 images

[CLEANED UP DATASET IS ON HUGGIN FACE](https://huggingface.co/datasets/Anarchiviste/GenHisDoc_dataset/tree/main)

GenHisDoc is a generalistic datasets for historical documents layout recognition and detection. GenHisDoc use a combination of several previously published datasets which have been adapted and re-annotated to work together and our own annotated data. All embedded datasets are licensed under Creative Commons or other licenses that permit reuse.

```markdown
GenHisDoc
├── images
├── labels
├── classes.txt
├── data.yaml # a YAML file with all the information needed to start a training session
├── origin.csv # Show you the dataset origin of a specific image.
├── split.py # split the dataset between a test, train and val with a seed of 42.
└── htmlgenerator.py # will create a html page with updated statistic about the dataset and a full annotation display on all images in the annotations/ folder.
```

The annotations are in YOLO format using the classes below. 
- Illustration
- Initial
- Ornament
- Stamp
- Table

## Published Datasets inside GenHisDoc

### The Sacrobosco Dataset (S-VED)
Paper : [CorDeep and the Sacrobosco Dataset: Detection of Visual Elements in Historical Documents](https://doi.org/10.3390/jimaging8100285)

Published by Jochen Büttner 1, Julius Martinetz 1,2, Hassan El-Hajj 1,2, Matteo Valleriani 1,2,3,4.

1 : Max Planck Institute for the History of Science, Boltzmannstr. 22, 14195 Berlin, Germany

2 : BIFOLD—Berlin Institute for the Foundations of Learning and Data, 10587 Berlin, Germany

3 : Institute of History and Philosophy of Science, Technology, and Literature, Faculty I—Humanities and Educational Sciences, Technische Universität Berlin, Straße des 17. Juni 135, 10623 Berlin, Germany

4 : The Cohn Institute for the History and Philosophy of Science and Ideas, Faculty of Humanities, Tel Aviv University, P.O. Box 39040, Ramat Aviv, Tel Aviv 6139001, Israel

Datasets : [The Sacrobosco Dataset](https://zenodo.org/record/7142456)

Modifications : The printer's mark annotation classe has been transfered to the illustration annotation classe. The format of the annotation in csv as been transformed into yolo style format with a txt attached to the image.

### IlluHisDoc

Paper : [docExtractor: An off-the-shelf historical document element extraction](https://arxiv.org/abs/2012.08191)

Published by Tom Monnier 1 et Mathieu Aubry 1

1 : LIGM, École nationale des Ponts et chaussées, Université Gustave Eiffel, CNRS, Marne-la-vallée, France

Datasets: [Illuhisdoc dropbox link](https://www.dropbox.com/scl/fi/ql0yxqapyyl0adbzzgn1x/illuhisdoc.zip?rlkey=q7mqkd3ljzwrk3lelkm2rgico&e=1&dl=0)

Modifications : Illuhisdoc use a per pixel segmentation with 4 classes, we transformed this segmentation in yolo style format detection.

### Horae LSv2

Datasets : [HORAE-LSv2. Layout Segmentation Dataset for Medieval Books of Hours (Version 2)](https://zenodo.org/records/16919911)

Published by Stutzmann Dominique 1, Bernard Leterme Lise 1, Boillet Mélodie 2, Bonhomme Marie-Laurence, Kermorvant Christopher 3

1 :  Institut de recherche et d'histoire des textes du Centre national de la recherche scientifique, Paris - Aubervilliers, 14, cours des Humanités, 93322 Aubervilliers

2 & 3 : Teklia, 30 rue Raymond Losserand, 75014 Paris, France

Modifications : Horae use a deep annotation system usefull only for manuscript, we reunited this classes into our segmentation ontology. We kept 4244 annotations about ornements, illustrations and initials, and suppressed 18720 annotations about text segmentation.

### El Picpic Stamps Dataset

The El PicPic stamps dataset was a CC by 4.0 licensed dataset on Roboflow, the origin of whose publication we cannot trace.

Published by the Roboflow user "El PicPic"

Modifications : A complete reannotation for stamps and an enrichement of annotation with tables.

## Unpublished Datasets inside GenHisDoc

### Aikon / Projet VHS / Eida 

[Aikon](https://aikon-platform.github.io/) is a IIIF automatic annotation project financed by the ERC project DISCOVER and developped between the [IMAGINE-LIGM laboratory at École nationale des ponts et chaussées](https://imagine-lab.enpc.fr/), [LTE laboratory at Observatoire de Paris-PSL](https://lte.observatoiredeparis.psl.eu/). Aikon is not a dataset, we aggregated and formated the open and human corrected annotated witness by the community in the Project VHS and Eida environnement of Aikon.

#### Projet VHS

 The ANR project [VHS](https://vhs.hypotheses.org/) is an interdisciplinary research project bringing together specialists in History of Science and Computer Vision to develop a new approach in the historical analysis of the circulation of scientific knowledge and the development of a visual scientific thought from the Middle Ages to the modern era, based on new methods of illustration analysis. 

Witness #2320 : Cyclopaedia, 5e éd., Vol. 2 - annotated by Alexandre

Witness #2365 : Latin 7416 | Paris, BnF - annotated by Alexandre

Witness #2377 : Cod. 44 | Österreichische Nationalbibliothek - annotated by Alexandre

Witness #2416 : Lat. Q. 9 | Universiteitsbibliotheek - annotated by Alexandre

Witness #2418 : Voss. Lat. Q. 40 | Universiteitsbibliotheek - annotated by Alexandre

Witness #2420 : 187 | Wien, Österreichische Nationalbibliothek - annotated by Alexandre

Witness #2421 : T. 47 | Biblioteca Ambrosiana - annotated by Alexandre

Witness #2387 : Latin 13955 | Paris, BnF - annotated by Alexandre 

Witness #2423 : Dc 183 | Dresden, Sächsische Landesbibliothek - annotated by Alexandre 

#### Projet EIDA

The project ANR EIDA aims at developing a new critique of astronomical diagrams of the 8th-18th centuries found in Chinese, Sanskrit, Persian/Arabic, Greek, Hebrew and Latin sources.  It is an interdisciplinary project joining a team of historians of sciences with computer vision researchers who will develop specific algorithms for retrieval and analysis of the historical diagrams. 

Witness #409 : Latin 7293A | Bibliothèque nationale de France (http://archivesetmanuscrits.bnf.fr/ark:/12148/cc66491c) annotated by sgessner

Witness #407 : Sprenger 1838 | Staatsbibliothek (https://dlc.mpg.de/!image/1762597527/10/-/) annotated by szieme

Witness #362 : Sanscrit 964 | Bibliothèque nationale de France (https://gallica.bnf.fr/ark:/12148/btv1b53119236k/f1) annotated by fsomer

Witness #360 : Supplément turc 242 | Bibliothèque nationale de France (https://gallica.bnf.fr/ark:/12148/btv1b8427189w) annotated by fsomer

Modifications : Complet correction and enrichement of Aikon mono-classe detection.

## Architecture of the project

[The architecture of the project](img/projectarchitecture.jpg)

[Classes distribution inside GenHisDoc](img/GenHisDoc_class_distribution.png)