GROBID is excellent software for recognizing various human-level objects in scientific papers (converting PDF into XML containing entities like table of contents, authors, titles, sections, chapters, bibliography, etc - this is so called TEI format).
However, it only works on PDF format.
I found in experiments that it generally works well on other PDFs, such as commercial research reports, e.g. reports published by Gartner or Forrester.
I need this type of conversion (sections, titles, list of authors, etc.) for other formats, such as HTML,
(In theory, it should be possible: commercial reports are often published both in HTML and PDF with the same document structure.)
Is there anything out there that could do the job like this?
Preferably, it should be open source like GROBID.