Hierarchical Cell-Level Parallelism for Large Tables
Process table cells in parallel during document parsing so complex HTML tables are reconstructed accurately from spatial coordinates.
Time
Compute level
Resources
Prerequisites
Scope
The Gap
The paper processes tables as single layout elements in Stage 2, generating the full HTML code autoregressively. The authors explicitly note in the limitations that 'there is potential for further optimization through parallel processing of... table cells.'
Limitation
Parsing large, complex tables as a single block creates extremely long token sequences for the autoregressive decoder. This increases latency and the risk of error propagation (hallucination) in dense numerical data.
Proposed Solution
Extend the Stage 1 layout analysis to hierarchically detect table rows and cells within table regions. In Stage 2, treat individual cells (or rows) as anchors for parallel recognition, then reconstruct the HTML structure deterministically based on spatial coordinates rather than generating it token-by-token.
Why This Matters
This would significantly accelerate the parsing of data-heavy financial or scientific documents and improve accuracy on massive tables where context windows might otherwise be exceeded.
How to Get Started
- Analyze the PubTabNet dataset to establish bounding box ground truth for individual table cells and rows.
- Extend the Stage 1 layout detection model to segment table regions into distinct cell-level components.
- Implement a parallel recognition module to process multiple cell crops simultaneously.
- Reconstruct the final HTML structure deterministically by mapping recognized text back to spatial coordinates.
Code linked to this paper on Hugging Face: github.com/bytedance/dolphin
Suggested by AI. Adapt it to your setup.
Has This Been Done?
None of the provided papers discuss hierarchical cell-level parallelism or deterministic spatial reconstruction for table layout analysis and structure recognition. Papers like Robust Table Detection and Structure Recognition from Heterogeneous Document Images and TSRFormer focus on standard transformer-based structure recognition rather than cell-level parallel anchoring and deterministic spatial HTML reconstruction.
- Robust Table Detection and Structure Recognition from Heterogeneous Document Images · 2022 · 102 citations
- TSRFormer: Table Structure Recognition with Transformers · 2022 · 94 citations
We searched OpenAlex for “Process table cells in parallel during document parsing so complex HTML tables are reconstructed accurately from spatial coordinates.” and “Hierarchical Cell-Level Parallelism for Large Tables” and an AI compared the idea with the top 6 results. A small search can miss relevant work.
Source Paper
Dolphin: Document Image Parsing via Heterogeneous Anchor PromptingHao Feng, Shu Wei, Xiang Fei +10 more · 2025
This paper introduces Dolphin, a multimodal document image parsing model that employs an "analyze-then-parse" paradigm to extract structured content from documents. By first identifying layout elements (e.g., paragraphs, tables) and their reading order, and then using these elements as anchors for parallel content parsing with specialized prompts, Dolphin achieves state-of-the-art performance and high efficiency compared to existing end-to-end VLMs.
Related Ideas
- Generative Vision-Language Model for Table RecognitionMinerU: An Open-Source Solution for Precise Document Content Extraction🔴Hard
- Multi-Granularity Visual Encoding for Small-Scale Document ElementsSmolDocling: An ultra-compact vision-language model for end-to-end multi-modal document conversion🟠Medium
- Direct Visual-to-Graph Generation for Irregular LayoutsRAG-Anything: All-in-One RAG Framework🟠Medium
- 2D-Attentional Adapters for Complex Table TopologyPaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model🔴Hard
- Dynamic Multi-Scale Layout AnalysisMinerU2.5: A Decoupled Vision-Language Model for Efficient High-Resolution Document Parsing🟡Fairly easy