README.md

---
license: cc-by-4.0
task_categories:
- image-segmentation
- object-detection
- mask-generation
version: "1.5.0"
pretty_name: CATMuS Medieval Segmentation
size_categories:
- 100K<n<1M
tags:
- layout-analysis
- humanities
- historical-documents
---
# Dataset Card for CATMuS Medieval (Segmentation Version)

![Banner for the CATMuS Project](banner_catmus_medieval_centered.png)

Join our Discord to ask questions about the dataset: [![Join the Discord](https://img.shields.io/badge/CATMuS-Discord?style=flat-square&logo=discord&logoColor=%23333333&color=%235865F2)](https://discord.gg/J38xgNEsGk)

## Dataset Details

**CATMuS Medieval Segmentation** (Consistent Approaches to Transcribing Manuscripts) is a specialized dataset designed for layout analysis of medieval manuscripts using the [SegmOnto](https://segmonto.github.io/) vocabulary for region and line classification. This dataset addresses the challenges associated with establishing consistent ground truth in layout analysis tasks, particularly for the complex and heterogeneous historical sources of medieval manuscripts in Latin scripts from the 8th to the 15th century CE. It is a subset of the manuscript present in the CATMuS Medieval dataset, which focuses on HTR only.

The CATMuS dataset for layout analysis provides:

- A uniform framework for annotation practices for the layout of medieval manuscripts.
- A benchmarking environment for evaluating automatic layout analysis models across multiple dimensions thanks to some metadata (for now, century of production).
- A benchmarking environment for other tasks (such as datation approaches).
- A platform for exploratory work in computer vision and digital paleography focused on layout-based tasks, such as layout generation.

Developed through collaboration among various institutions and projects, CATMuS Medieval offers an inter-compatible dataset that spans over 200 manuscripts and incunabula in 10 different languages, containing a wealth of structural annotations using the SegmOnto vocabulary.

By ensuring consistency in layout analysis approaches, CATMuS aims to mitigate challenges arising from the diversity in standards for medieval manuscript analysis. It provides a comprehensive benchmark for evaluating layout analysis models on historical sources, facilitating advancements in the field of digital humanities.

### Dataset Description

<!-- Provide a longer summary of what this dataset is. -->

- **Curated by:** Thibault Clérice (Inria)
- **Funded by:** BnF Datalab, Biblissima +, DIM PAMIR
- **License:** CC-BY 4.0

#### Documents

|                               |   train |   dev |   test |   Total |
|:------------------------------|--------:|------:|-------:|--------:|
| images                        |    1336 |   191 |    178 |    1705 |
| manuscripts                   |     159 |    20 |     28 |     207 |

#### Century coverage

*As the number of images in each split. Images can represent two pages.*

|                               |   train |   dev |   test |   Total |
|:------------------------------|--------:|------:|-------:|--------:|
| Century:08                    |       2 |     0 |      0 |       2 |
| Century:09                    |     111 |     1 |      0 |     112 |
| Century:10                    |      11 |     0 |     38 |      49 |
| Century:11                    |      27 |     0 |      0 |      27 |
| Century:12                    |      19 |    17 |     10 |      46 |
| Century:13                    |     230 |     9 |     20 |     259 |
| Century:14                    |     241 |   111 |     39 |     391 |
| Century:15                    |     563 |    36 |     19 |     618 |
| Century:16                    |     132 |    17 |     52 |     201 |

#### Lines

|                               |   train |   dev |   test |   Total |
|:------------------------------|--------:|------:|-------:|--------:|
| Line:DefaultLine              |   81783 | 13554 |  12595 |  107932 |
| Line:DropCapitalLine          |    1175 |   105 |    100 |    1380 |
| Line:HeadingLine              |    1381 |   701 |    165 |    2247 |
| Line:InterlinearLine          |    2808 |    27 |   2234 |    5069 |
| Line:MusicLine                |     167 |     0 |      0 |     167 |
| Line:TironianSignLine         |     282 |     0 |      0 |     282 |

#### Zones

|                               |   train |   dev |   test |   Total |
|:------------------------------|--------:|------:|-------:|--------:|
| Zone:DamageZone               |      12 |     1 |      0 |      13 |
| Zone:DigitizationArtefactZone |      28 |     0 |      0 |      28 |
| Zone:DropCapitalZone          |    1567 |   102 |    132 |    1801 |
| Zone:GraphicZone              |     300 |     7 |     15 |     322 |
| Zone:MainZone                 |    2317 |   365 |    294 |    2976 |
| Zone:MarginTextZone           |     916 |   146 |    199 |    1261 |
| Zone:MusicZone                |     179 |     0 |      0 |     179 |
| Zone:NumberingZone            |     632 |   102 |     95 |     829 |
| Zone:QuireMarksZone           |      86 |     9 |     15 |     110 |
| Zone:RunningTitleZone         |     340 |    91 |     18 |     449 |
| Zone:SealZone                 |       3 |     0 |      0 |       3 |
| Zone:StampZone                |      39 |     5 |      5 |      49 |
| Zone:TitlePageZone            |       4 |     1 |      2 |       7 |


## Uses

### Direct Use

- Layout Analysis

### Out-of-Scope Use

- Text-To-Image

## Dataset Structure

- `data` contains 3 splits, which are loaded through `load_dataset("CATMuS/medieval-segmentation")`. They are the same split as [Catmus Medieval (for HTR)](https://huggingface.co/datasets/CATMuS/medieval)
- Each image is annotated with a
  - `file_name` (path from root)
  - `shelfmark` identifier
  - `century` datation information
  - `project` that originally produced the data 
  - `width` of the page (in pixels)
  - `height` of the page (in pixels)
  - `objects` which contains sequences of values for each object found in the page:
    - `id` that are mainly used for parent relationship between blocks (such as columns) and lines
    - `bbox` (Shape: `[x1, y1, x2, y2]`, top left to bottom right)
    - `polygons` (Shape: `[x, y, x, y, x, y, ...]`)
    - `category` as a list of string using the first level of [SegmOnto](https://segmonto.github.io/) guidelines
    - `type` which is either `block` (area) or `line`.
    - `parent` which hold the id of the parent (is null for blocks, is nullable for line)

<!--

## Dataset Creation

### Curation Rationale

 Motivation for the creation of this dataset.

[More Information Needed]

### Source Data

 This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). 

#### Data Collection and Processi

#### Who are the source data producers?

 -->

### Annotations

#### Annotation process

The annotation process is described in the [dataset paper](https://inria.hal.science/hal-04453952).

#### Who are the annotators?

- Pinche, Ariane
- Clérice, Thibault
- Chagué, Alix
- Camps, Jean-Baptiste
- Vlachou-Efstathiou, Malamatenia
- Gille Levenson, Matthias
- Brisville-Fertin, Olivier
- Boschetti, Federico
- Fischer, Franz
- Gervers, Michael
- Boutreux, Agnès
- Manton, Avery
- Gabay, Simon
- Bordier, Julie
- Glaise, Anthony
- Alba, Rachele
- Rubin, Giorgia
- White, Nick
- Karaisl, Antonia
- Leroy, Noé
- Maulu, Marco
- Biay, Sébastien
- Cappe, Zoé
- Konstantinova, Kristina
- Boby, Victor
- Christensen, Kelly
- Pierreville, Corinne
- Aruta, Davide
- Lenzi, Martina
- Le Huëron, Armelle
- Possamaï, Marylène
- Duval, Frédéric
- Mariotti, Violetta
- Morreale, Laura
- Nolibois, Alice
- Foehr-Janssens, Yasmina
- Deleville, Prunelle
- Carnaille, Camille
- Lecomte, Sophie
- Meylan, Aminoel
- Ventura, Simone
- Dugaz, Lucien

#### Software

The software to generate this version of the dataset was built by Thibault Clérice and William Mattingly

## Bias, Risks, and Limitations

The data are skewed toward Old French, Middle Dutch and Spanish, specifically from the 14th century.

The only language that is represented over all centuries is Latin, and in each scripts. The other language with a coverage close to Latin is Old French.

Only one document is available in Old English.

## Citation

**BibTeX:**

```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},
}
```

**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⟩


## Dataset Card Contact

Thibault Clérice (first.last@inria.fr)