generate_bbox_data.py

from pathlib import Path

import openparse

# directory for sample PDF's
PDF_DIR = Path("../data") / "pdfs"

# output dir
GT_DIR = Path("../data") / "ground_truth"

parser = openparse.DocumentParser()

for pdf_path in PDF_DIR.glob("*.pdf"):
    parsed_doc = parser.parse(pdf_path)

    with open(GT_DIR / pdf_path.name.replace(".pdf", ".json"), "w") as f:
        f.write(parsed_doc.model_dump_json())

    pdf = openparse.Pdf(pdf_path)
    
    ### === OPTIONAL: display nodes === ### 
    # useful for filtering out "bad" nodes since open-parse is quite simple in how it parses
    for node in parsed_doc.nodes:
        print(node.text)

        pdf.display_with_bboxes([node], page_nums=[node.start_page])