dvgodoy/ICDAR2019_cTDaR_TRACKB_resized

Dataset Card for ICDAR2019-cTDaR-TRACKB

This dataset is a resized version of the original cndplab-founder/ICDAR2019_cTDaR, merged with with its supplement cndplab-founder/ICDAR2019_cTDaR_dataset_supplement.

You can easily and quickly load it:

dataset = load_dataset("dvgodoy/ICDAR2019_cTDaR_TRACKB_resized")

DatasetDict({
    train: Dataset({
        features: ['image', 'width', 'height', 'category', 'label', 'bboxes_table', 'bboxes_cell'],
        num_rows: 1200
    })
    test: Dataset({
        features: ['image', 'width', 'height', 'category', 'label', 'bboxes_table', 'bboxes_cell'],
        num_rows: 390
    })
})

Dataset Summary

From the original ICDAR2019 cTDaR dataset page:

The dataset consists of modern documents and archival ones with various formats, including document images and born-digital formats such as PDF. The annotated contents contain the table entities and cell entities in a document, while we do not deal with nested tables.

This "resized" version contains all the images from "Track B" (table recognition) resized so that the largest dimension (either width or height) is 1000px. The annotations were converted from XML to JSON and boxes are represented in Pascal VOC format (xmin, ymin, xmax, ymax).

For the modern dataset no training data is available for Track B.

The original dataset did not contain "modern" tables or annotations for "Track B", so the supplement dataset was merged into it, and its annotations converted accordingly.

Dataset Structure

Data Instances

A sample from the training set is provided below :

{
    'image': <PIL.JpegImagePlugin.JpegImageFile image mode=L size=1000x729>,
    'width': 1000,
    'height': 729,
    'category': 'historical',
    'label': 0,
    'bboxes_table': [[...]],
    'bboxes_cell': [[...]]
}

Data Fields

• image: A PIL.Image.Image object containing a document.
• width: image's width.
• height: image's height.
• category: class label.
• label: an int classification label.
• bboxes_table: list of box coordinates in (xmin, ymin, xmax, ymax) format (Pascal VOC).
• bboxes_cell: list of lists of box coordinates in (xmin, ymin, xmax, ymax) format (Pascal VOC) - the outer list matches the length of the bboxes_table list, and each of its elements is a list of cells.

Class Label Mappings

{
  "0": "historical",
  "1": "modern"
}

Data Splits

	train	test
# of examples	1200	390

Additional Information

Licensing Information

This dataset is a resized and reorganized version of ICDAR2019 cTDaR from the ICDAR 2019 Competition on Table Detection and Recognition, merged with its supplement, which is licensed under BSD 2-Clause License.