What problem does it solve?
Converts complex PDFs and document images into complete, structured Markdown/JSON with layout-aware extraction so users no longer need to manually reconstruct tables, formulas, multi-column text, figures, stamps, or reading order from scans and print documents.
Core Features & Use Cases
- Table extraction: Recover table structure and cell contents for invoices, financial reports, and spreadsheets.
- Formula recognition: Detect and output mathematical formulas as LaTeX for academic papers and technical documents.
- Layout analysis & reading order: Restore multi-column layouts, headers/footers, page numbers, and correct reading order for magazines, newspapers, and brochures.
- Figures, charts, seals: Identify figures, charts, and stamps and include associated metadata and image references.
- Use Case: Parse a scanned annual report to extract all text, tables, and formulas into Markdown and export raw provider JSON for downstream data pipelines.
Quick Start
Run the vl_caller.py script after configuring PADDLEOCR_DOC_PARSING_API_URL and PADDLEOCR_ACCESS_TOKEN and provide a file URL or local file path to receive the full structured JSON and markdown output.