biglam/yalta_ai_segmonto_manuscript_dataset

YALTAi Segmonto Manuscript and Early Printed Book Dataset

Dataset Summary

This dataset contains a subset of data used in the paper You Actually Look Twice At it (YALTAi): using an object detection approach instead of region segmentation within the Kraken engine. This paper proposes treating page layout recognition on historical documents as an object detection task (compared to the usual pixel segmentation approach). This dataset contains images from digitised manuscripts and early printed books with the following labels:

• DamageZone
• DigitizationArtefactZone
• DropCapitalZone
• GraphicZone
• MainZone
• MarginTextZone
• MusicZone
• NumberingZone
• QuireMarksZone
• RunningTitleZone
• SealZone
• StampZone
• TableZone
• TitlePageZone

Supported Tasks and Leaderboards

• object-detection: This dataset can be used to train a model for object-detection on historic document images.

Dataset Structure

This dataset has two configurations. These configurations both cover the same data and annotations but provide these annotations in different forms to make it easier to integrate the data with existing processing pipelines.

• The first configuration, YOLO, uses the data's original format.
• The second configuration converts the YOLO format into a format closer to the COCO annotation format. This is done to make it easier to work with the feature_extractor from the Transformers models for object detection, which expect data to be in a COCO style format.

Data Instances

An example instance from the COCO config:

{'height': 5610,
 'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=3782x5610 at 0x7F3B785609D0>,
 'image_id': 0,
 'objects': [{'area': 203660,
   'bbox': [1545.0, 207.0, 1198.0, 170.0],
   'category_id': 9,
   'id': 0,
   'image_id': '0',
   'iscrowd': False,
   'segmentation': []},
  {'area': 137034,
   'bbox': [912.0, 1296.0, 414.0, 331.0],
   'category_id': 2,
   'id': 0,
   'image_id': '0',
   'iscrowd': False,
   'segmentation': []},
  {'area': 110865,
   'bbox': [2324.0, 908.0, 389.0, 285.0],
   'category_id': 2,
   'id': 0,
   'image_id': '0',
   'iscrowd': False,
   'segmentation': []},
  {'area': 281634,
   'bbox': [2308.0, 3507.0, 438.0, 643.0],
   'category_id': 2,
   'id': 0,
   'image_id': '0',
   'iscrowd': False,
   'segmentation': []},
  {'area': 5064268,
   'bbox': [949.0, 471.0, 1286.0, 3938.0],
   'category_id': 4,
   'id': 0,
   'image_id': '0',
   'iscrowd': False,
   'segmentation': []},
  {'area': 5095104,
   'bbox': [2303.0, 539.0, 1338.0, 3808.0],
   'category_id': 4,
   'id': 0,
   'image_id': '0',
   'iscrowd': False,
   'segmentation': []}],
 'width': 3782}

An example instance from the YOLO config:

{'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=3782x5610 at 0x7F3B785EFA90>,
 'objects': {'bbox': [[2144, 292, 1198, 170],
   [1120, 1462, 414, 331],
   [2519, 1050, 389, 285],
   [2527, 3828, 438, 643],
   [1593, 2441, 1286, 3938],
   [2972, 2444, 1338, 3808]],
  'label': [9, 2, 2, 2, 4, 4]}}

Data Fields

The fields for the YOLO config:

• image: the image
• objects: the annotations which consist of:
  • bbox: a list of bounding boxes for the image
  • label: a list of labels for this image

The fields for the COCO config:

• height: height of the image
• width: width of the image
• image: image
• image_id: id for the image
• objects: annotations in COCO format, consisting of a list containing dictionaries with the following keys:
  • bbox: bounding boxes for the images
  • category_id: a label for the image
  • image_id: id for the image
  • iscrowd: COCO is a crowd flag
  • segmentation: COCO segmentation annotations (empty in this case but kept for compatibility with other processing scripts)

Data Splits

The dataset contains a train, validation and test split with the following numbers per split:

Dataset	Number of images
Train	854
Dev	154
Test	139

A more detailed summary of the dataset (copied from the paper):

	Train	Dev	Test	Total	Average area	Median area
DropCapitalZone	1537	180	222	1939	0.45	0.26
MainZone	1408	253	258	1919	28.86	26.43
NumberingZone	421	57	76	554	0.18	0.14
MarginTextZone	396	59	49	504	1.19	0.52
GraphicZone	289	54	50	393	8.56	4.31
MusicZone	237	71	0	308	1.22	1.09
RunningTitleZone	137	25	18	180	0.95	0.84
QuireMarksZone	65	18	9	92	0.25	0.21
StampZone	85	5	1	91	1.69	1.14
DigitizationArtefactZone	1	0	32	33	2.89	2.79
DamageZone	6	1	14	21	1.50	0.02
TitlePageZone	4	0	1	5	48.27	63.39

Dataset Creation

This dataset is derived from:

• CREMMA Medieval ( Pinche, A. (2022). Cremma Medieval (Version Bicerin 1.1.0) Data set
• CREMMA Medieval Lat (Clérice, T. and Vlachou-Efstathiou, M. (2022). Cremma Medieval Latin Data set
• Eutyches. (Vlachou-Efstathiou, M. Voss.Lat.O.41 - Eutyches "de uerbo" glossed Data set
• Gallicorpora HTR-Incunable-15e-Siecle ( Pinche, A., Gabay, S., Leroy, N., & Christensen, K. Données HTR incunable du 15e siècle Computer software
• Gallicorpora HTR-MSS-15e-Siecle ( Pinche, A., Gabay, S., Leroy, N., & Christensen, K. Données HTR manuscrits du 15e siècle Computer software
• Gallicorpora HTR-imprime-gothique-16e-siecle ( Pinche, A., Gabay, S., Vlachou-Efstathiou, M., & Christensen, K. HTR-imprime-gothique-16e-siecle Computer software

• a few hundred newly annotated data, specifically the test set which is completely novel and based on early prints and manuscripts.

These additional annotations were created by correcting an early version of the model developed in the paper using the roboflow platform.

Curation Rationale

[More information needed]

Source Data

The sources of the data are described above.

Initial Data Collection and Normalization

[More information needed]

Who are the source language producers?

[More information needed]

Annotations

Annotation process

Additional annotations produced for this dataset were created by correcting an early version of the model developed in the paper using the roboflow platform.

Who are the annotators?

[More information needed]

Personal and Sensitive Information

This data does not contain information relating to living individuals.

Considerations for Using the Data

Social Impact of Dataset

A growing number of datasets are related to page layout for historical documents. This dataset offers a different approach to annotating these datasets (focusing on object detection rather than pixel-level annotations). Improving document layout recognition can have a positive impact on downstream tasks, in particular Optical Character Recognition.

Discussion of Biases

Historical documents contain a wide variety of page layouts. This means that the ability of models trained on this dataset to transfer to documents with very different layouts is not guaranteed.

Other Known Limitations

[More information needed]

Additional Information

Dataset Curators

Licensing Information

Creative Commons Attribution 4.0 International

Citation Information

@dataset{clerice_thibault_2022_6814770,
  author       = {Clérice, Thibault},
  title        = {{YALTAi: Segmonto Manuscript and Early Printed Book 
                   Dataset}},
  month        = jul,
  year         = 2022,
  publisher    = {Zenodo},
  version      = {1.0.0},
  doi          = {10.5281/zenodo.6814770},
  url          = {https://doi.org/10.5281/zenodo.6814770}
}

Contributions

Thanks to @davanstrien for adding this dataset.