Comment by djprice1

What do you mean when you say "vector geometry"? Are you using the geometry extracted from PDFs directly? I'm curious how that interacts with the OCR and detection model portion of what you're doing