ocr-and-documents

Use pymupdf to open the PDF and call get_text() on each page, or run the extract_pymupdf.py script with the file path. For markdown output, install pymupdf4llm and use the --markdown flag.

Use pymupdf for text-based PDFs since it is only ~25MB and runs instantly. Use marker-pdf when you need OCR for scanned documents, equation extraction, forms, or complex layout analysis, but it requires ~3-5GB for PyTorch and models.

No, pymupdf cannot perform OCR on scanned documents since it only reads existing text layers. For scanned PDFs, use marker-pdf, which supports OCR in over 90 languages.

Run extract_pymupdf.py with the --tables flag, which uses pymupdf's find_tables() to detect tables and outputs them as markdown via pandas. For higher accuracy on complex tables, marker-pdf provides better results.

marker-pdf requires ~3-5GB of disk space for PyTorch and downloads ~2.5GB of models to ~/.cache/huggingface/ on first use. Run extract_marker.py --check first to verify you have at least 5GB free.

Use web_extract with the document URL, which converts PDFs to markdown via Firecrawl without any local dependencies. Only fall back to local extraction with pymupdf or marker-pdf if web_extract fails or the file is local.