Hi, I think our approach can benefit from these OCR tools to generate a better tree for search. We will update more on this point in our GitHub repo, thanks for raising this question!