Documents can have complex structures, like multi-column text blocks and floating figures. Older OCR models handled these documents by detecting words and then the layout of pages manually in ...