document layout analysis: the PubLayNet dataset (a collection of more
  than 360,000 document images constructed by automatically parsing PubMed XML files).