pdf

1#!/usr/bin/env python32"""Extract text from documents using marker-pdf. High-quality OCR + layout analysis.3 4Requires ~3-5GB disk (PyTorch + models downloaded on first use).5Supports: PDF, DOCX, PPTX, XLSX, HTML, EPUB, images.6 7Usage:8    python extract_marker.py document.pdf9    python extract_marker.py document.pdf --output_dir ./output10    python extract_marker.py presentation.pptx11    python extract_marker.py spreadsheet.xlsx12    python extract_marker.py scanned_doc.pdf           # OCR works here13    python extract_marker.py document.pdf --json        # Structured output14    python extract_marker.py document.pdf --use_llm     # LLM-boosted accuracy15"""16import sys17import os18 19def convert(path, output_dir=None, output_format="markdown", use_llm=False):20    from marker.converters.pdf import PdfConverter21    from marker.models import create_model_dict22    from marker.config.parser import ConfigParser23 24    config_dict = {}25    if use_llm:26        config_dict["use_llm"] = True27 28    config_parser = ConfigParser(config_dict)29    models = create_model_dict()30    converter = PdfConverter(config=config_parser.generate_config_dict(), artifact_dict=models)31    rendered = converter(path)32 33    if output_format == "json":34        import json35        print(json.dumps({36            "markdown": rendered.markdown,37            "metadata": rendered.metadata if hasattr(rendered, "metadata") else {},38        }, indent=2, ensure_ascii=False))39    else:40        print(rendered.markdown)41 42    # Save images if output_dir specified43    if output_dir and hasattr(rendered, "images") and rendered.images:44        from pathlib import Path45        Path(output_dir).mkdir(parents=True, exist_ok=True)46        for name, img_data in rendered.images.items():47            img_path = os.path.join(output_dir, name)48            with open(img_path, "wb") as f:49                f.write(img_data)50        print(f"\nSaved {len(rendered.images)} image(s) to {output_dir}/", file=sys.stderr)51 52 53def check_requirements():54    """Check disk space before installing."""55    import shutil56    free_gb = shutil.disk_usage("/").free / (1024**3)57    if free_gb < 5:58        print(f"⚠️  Only {free_gb:.1f}GB free. marker-pdf needs ~5GB for PyTorch + models.")59        print("Use pymupdf instead (scripts/extract_pymupdf.py) or free up disk space.")60        sys.exit(1)61    print(f"✓ {free_gb:.1f}GB free — sufficient for marker-pdf")62 63 64if __name__ == "__main__":65    import argparse66 67    parser = argparse.ArgumentParser(68        description="Extract text from documents using marker-pdf (high-quality OCR + layout analysis)."69    )70    parser.add_argument("path", nargs="?", help="Document to convert (PDF, DOCX, PPTX, XLSX, HTML, EPUB, image)")71    parser.add_argument("--output_dir", help="Directory to save extracted images")72    parser.add_argument("--json", action="store_true", help="Structured JSON output instead of markdown")73    parser.add_argument("--use_llm", action="store_true", help="LLM-boosted accuracy")74    parser.add_argument("--check", action="store_true", help="Check disk space requirements and exit")75    args = parser.parse_args()76 77    if args.check:78        check_requirements()79        sys.exit(0)80    if not args.path:81        parser.error("path is required unless --check is given")82 83    convert(84        args.path,85        output_dir=args.output_dir,86        output_format="json" if args.json else "markdown",87        use_llm=args.use_llm,88    )89