Datasets:

Languages:
English
License:
historical_dla / README.md
jnaiman's picture
Update README.md
70e421a verified
metadata
license: apache-2.0
language:
  - en

Dataset Introduction

This dataset has bounding boxes for ~6000 hand annotated pages with bounding boxes for figures, figure captions, tables, and math formulas.

More coverage is available for figures + captions (some pages might not have all tables and math formulas annotated).

Format is JSON and includes lists (so it looks like HuggingFace doesn't necessarily like this format for display), with rows that look like:

image/png

How to use this data

To plot an example check out the trial_example_from_data.ipynb notebook.

This assumes you have the data and the data_utils.py file in the same location as your notebook.

The following packages will have to be installed:

matplotlib
numpy
pandas
wand
PIL
wget
cv2 # OpenCV

On Google Colab, to install wand we found we had to do the following (this is not in the linked notebook):

!apt install imagemagick
!apt-get install libmagickwand-dev

!pip install Wand

!rm /etc/ImageMagick-6/policy.xml

!pip install wget