LingDT-VL-OCR: Structure-Aware Document-Level Parsing with Fine-Grained Visual Reference

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, FinTech & Digital Financial Services · Depth: Expert, extended

Summary

Agentar-Fin-OCR is a document parsing system developed by Ant Group, specifically designed for financial-domain documents. It transforms ultra-long financial PDFs into semantically consistent, highly accurate, structured outputs with auditing-grade provenance. The system addresses challenges like complex layouts and cross-page structural discontinuities by integrating a Cross-page Contents Consolidation algorithm and a Document-level Heading Hierarchy Reconstruction (DHR) module. For precise table parsing, it employs a difficulty-adaptive curriculum learning strategy and a CellBBoxRegressor that localizes table cells using structural anchor tokens. The authors also introduce FinDocBench, a new benchmark comprising six financial document categories with expert-verified annotations and specialized metrics like TocEDS and C-IoU. Agentar-Fin-OCR demonstrates state-of-the-art performance in table parsing on OmniDocBench v1.5, achieving a Table^TEDS score of 92.82, and shows superior results on FinDocBench for layout detection and heading hierarchy reconstruction.

Key takeaway

For AI Scientists and ML Engineers developing document intelligence solutions for financial institutions, you should prioritize document-level parsing systems like Agentar-Fin-OCR. This approach overcomes semantic fragmentation in ultra-long financial PDFs, ensuring auditing-grade provenance and accurate information extraction, which is critical for compliance and reliable RAG applications. Utilize FinDocBench to rigorously evaluate your models against real-world financial document complexities and achieve superior table parsing and hierarchy reconstruction.

Key insights

Agentar-Fin-OCR provides auditing-grade, document-level parsing for complex financial PDFs, ensuring semantic and structural continuity across pages.

Principles

• Document-level parsing restores global context.
• Visual cues enhance hierarchy reconstruction.
• Difficulty-adaptive training improves table parsing.

Method

Agentar-Fin-OCR uses Cross-page Contents Consolidation and Document-level Heading Hierarchy Reconstruction. It applies curriculum learning with Group Relative Policy Optimization for table parsing and a CellBBoxRegressor for cell-level bounding box regression from decoder hidden states.

In practice

• Use FinDocBench for financial document evaluation.
• Implement cross-page content merging for long PDFs.
• Ground table cells for audit traceability.

Topics

• Document Parsing
• Financial Document Analysis
• Table Structure Recognition
• Heading Hierarchy Reconstruction
• FinDocBench
• Cell-Level Grounding

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• Article: Redesigning Banking PDF Table Extraction: A Layered Approach with Java — Reliable PDF table extraction in finance requires a layered architectural approach, not just a single parsing library.
• Deep Dive into Semantic Formatting Score: A New Metric for Meaningful Document Formatting — Document OCR benchmarks must evaluate semantic formatting for accurate AI agent interpretation.
• How to Build Your Own (DIY) Document Parsing Agent from Scratch — Agentic systems enhance GenAI document parsing by addressing VLM limitations like token limits and repetition.
• From Flutter On-Device to Gemini Vision: Building a Multi-Language Document Extraction Engine — Hybrid Vision LLM and OCR excels at structured data extraction from diverse, unstructured documents by combining visual and numeric accuracy.
• Benchmarking RAG Architectures Locally on a Real Financial PDF — Local RAG on complex financial PDFs requires specialized extraction and architectural choices to achieve meaningful accuracy.
• MPDocBench-Parse: Benchmarking Practical Multi-page Document Parsing — Practical multi-page document parsing requires benchmarks assessing semantic continuity, visual content, and hierarchical structure beyond basic text.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.