PubLayNet

Format	JPG images with JSON label
License	CLDA-Permissive
Domain	Computer Vision
Number of Records	340,000 images
Size	102 GB
Data Origin	Images of research papers from PubMed and annotations from IBM Research Australia.
Dataset Version	Version 1 - August 07, 2019
Dataset Coverage	The dataset contains images of research papers from the medical domain.
Business Use Case	Document Understanding The dataset can be used to train a model to extract various elements of a document such as tables, figures, texts etc. This can aid businesses dealing with a large number of documents to easily categorize the various elements in their documents.

Image Annotated Image (Generated with Notebook) JSON

Image	Annotated Image (Generated with Notebook)	JSON
		{ "images": [ { "file_name": "PMC5491943_00004.jpg", "height": 794, "id": 348952, "width": 596 }, { "file_name": "PMC5302692_00002.jpg", "height": 792, "id": 384435, "width": 612 }, { "file_name": "PMC3863500_00003.jpg", "height": 792, "id": 353156, "width": 601 } ], "annotations": [ { "segmentation": [ [ 37.59, 360.34, 288.66, 360.34, 288.66, 370.34, 288.63, 370.34, 288.63, 381.37, 288.63, 381.37, 288.63, 391.26, 272.86, 391.26, 272.86, 401.7, 37.59, 401.7, 37.59, 391.8, 37.59, 381.37, 37.59, 370.89, 37.59, 360.34 ] ], "area": 10218.471181684348, "iscrowd": 0, "image_id": 346767, "bbox": [ 37.59, 360.34, 251.07, 41.36 ], "category_id": 1, "id": 3377124 } ] }

{
  "images": [
    {
      "file_name": "PMC5491943_00004.jpg",
      "height": 794,
      "id": 348952,
      "width": 596
    },
    {
      "file_name": "PMC5302692_00002.jpg",
      "height": 792,
      "id": 384435,
      "width": 612
    },
    {
      "file_name": "PMC3863500_00003.jpg",
      "height": 792,
      "id": 353156,
      "width": 601
    }
  ],
  "annotations": [
    {
      "segmentation": [
        [
          37.59,
          360.34,
          288.66,
          360.34,
          288.66,
          370.34,
          288.63,
          370.34,
          288.63,
          381.37,
          288.63,
          381.37,
          288.63,
          391.26,
          272.86,
          391.26,
          272.86,
          401.7,
          37.59,
          401.7,
          37.59,
          391.8,
          37.59,
          381.37,
          37.59,
          370.89,
          37.59,
          360.34
        ]
      ],
      "area": 10218.471181684348,
      "iscrowd": 0,
      "image_id": 346767,
      "bbox": [
        37.59,
        360.34,
        251.07,
        41.36
      ],
      "category_id": 1,
      "id": 3377124
    }
  ]
}

Feature	Description
images	JSON field containing a list of images and their metadata (size, ID, name)
annotations	Each object instance annotation contains a series of fields, including the category id and segmentation mask of the object.
annotations -> segmentations	Contains the polygon coordinates for the segmentation mask for the specific class instance (table, list, text etc)
annotations -> bbox	Contains the bounding box coordinates for the specific class instance (table, list, text etc).
annotations -> is_crowd	This field indicates whether the class instance is a single object (is_crowd=0) or multiple objects (is_crowd=1). In this dataset we only have single objects so this field is always set to 0.
annotations -> category_id	The class label for the current class instance. This indicates what the current bbox/segmentation mask encapsulates (table, list, text etc).
categories	JSON field containing a list of classes and their metadata (ID, name) This dataset has 5 categories (w/ corresponding "ids") - text ("1"), title ("2"), list ("3"), table ("4"), figure ("5").