Datasets:

ds4sd

/

DocLayNet-v1.1

like 4

Some cells are misaligned, or I don't know what is wrong with them. Take into example the following row:

dataset["train"][1]
"""
{
  ...
  "pdf_cells": [
    [
      {
        "bbox": [
          159.4759894151565,
          712.7215975378788,
          31.568154324546953,
          14.008320391414145
        ],
        "font": {
          "color": [
            240,
            161,
            56,
            255
          ],
          "name": "/Avenir-Roman",
          "size": 1.0
        },
        "text": "Cor"
      },
      {
        "bbox": [
          543.8887928336079,
          712.7215975378788,
          13.76002553542014,
          14.008320391414145
        ],
        "font": {
          "color": [
            240,
            161,
            56,
            255
          ],
          "name": "/Avenir-Roman",
          "size": 1.0
        },
        "text": "Global Star franchises include the hit"
      },
    ...
    ],
    []
  ],
  "metadata": {
    "coco_height": 1025,
    "coco_width": 1025,
    "collection": "ann_reports_00_04_fancy",
    "doc_category": "financial_reports",
    "image_id": 1,
    "num_pages": 54,
    "original_filename": "NASDAQ_TTWO_2003.pdf",
    "original_height": 792.0,
    "original_width": 607.0,
    "page_hash": "f446422ed85e300319d3aff929762b8527cbc9ec26f55703b36df3e7447cc2d5",
    "page_no": 5
  }
}
"""

Is it intentionally? Where can I find original PDFs dataset instead of this COCO mess? Found somewhat original files, and dataset is named "extra", whatever extra it is.