PDF to Markdown

[nglong14]

I want to convert PDF to Markdown for my chatbot training.
Input PDFs (text based or image based) then convert to Markdown.

[github-actions]

Thank you for raising this issue! Our team will review it soon.

For more queries or discussions or approval of requests, please join the Discord server for further discussion: Discord Link

[chao-dotcom]

Hi! I joined the Discord but it seems inactive, so I’ll start working on a minimal implementation for this feature.

I plan to begin with text-based PDF → Markdown conversion, keeping the scope small and clean.

If this aligns with the project direction, I’ll open a PR for review. Happy to adjust based on feedback.

[EddDoubleD]

Exaple for docling + ocrmypdf

import argparse
from pathlib import Path
import ocrmypdf

from docling.document_converter import DocumentConverter


def run_ocr(input_pdf: Path, ocr_pdf: Path, lang: str) -> None:
    ocrmypdf.ocr(
        input_file=str(input_pdf),
        output_file=str(ocr_pdf),
        language=lang,
        deskew=True,
        rotate_pages=True,
        optimize=0,
        skip_text=True,
        progress_bar=False,
    )


def convert_one(pdf: Path, out_dir: Path, converter: DocumentConverter, lang: str, keep_ocr_pdf: bool):
    out_dir.mkdir(parents=True, exist_ok=True)

    ocr_pdf = out_dir / f"{pdf.stem}.ocr.pdf"
    md_path = out_dir / f"{pdf.stem}.md"

    # OCR -> PDF with text layer
    run_ocr(pdf, ocr_pdf, lang)

    # docling -> markdown
    result = converter.convert(str(ocr_pdf))
    md = result.document.export_to_markdown()
    md_path.write_text(md, encoding="utf-8")

    if not keep_ocr_pdf and ocr_pdf.exists():
        ocr_pdf.unlink()

    return md_path


def main():
    ap = argparse.ArgumentParser("Batch PDF -> OCR -> docling -> Markdown")
    ap.add_argument("in_dir", type=Path, help="PDF folder")
    ap.add_argument("-o", "--out-dir", type=Path, default=None, help="folder for .md (по умолчанию <in_dir>/md)")
    ap.add_argument("--lang", default="rus+eng", help='Языки Tesseract, напр. "rus+eng"')
    ap.add_argument("--recursive", action="store_true", help="Искать PDF рекурсивно")
    ap.add_argument("--keep-ocr-pdf", action="store_true", help="Сохранять промежуточные *.ocr.pdf")
    args = ap.parse_args()

    in_dir = args.in_dir
    out_dir = args.out_dir or (in_dir / "md")

    pdfs = sorted(in_dir.rglob("*.pdf") if args.recursive else in_dir.glob("*.pdf"))
    if not pdfs:
        print("PDF not found")
        return

    converter = DocumentConverter()

    ok, fail = 0, 0
    for pdf in pdfs:
        try:
            rel_parent = pdf.parent.relative_to(in_dir) if args.recursive else Path(".")
            target_dir = out_dir / rel_parent

            md_path = convert_one(pdf, target_dir, converter, args.lang, args.keep_ocr_pdf)
            print(f"[OK]  {pdf} -> {md_path}")
            ok += 1
        except Exception as e:
            print(f"[ERR] {pdf}: {e}")
            fail += 1

    print(f"Done. OK={ok}, ERR={fail}")


if __name__ == "__main__":
    main()