Research map / Vision-language and multimodal / Human and embodied modeling
Image-based table recognition: data, model, and evaluation
Xu Zhong, Elaheh ShafieiBavani, Antonio Jimeno Yepes
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use dual decoders (structure + cell) instead of a single decoder (EDD vs WYGIWYS) | EDD · PubTabNet test (all tables) | average TEDS (%): 78.6 -> 88.3 | Improved |
| use dual decoders on complex tables | EDD · PubTabNet test (complex tables) | average TEDS (%): 75.5 -> 85.4 | Improved |
| use dual decoders on simple tables | EDD · PubTabNet test (simple tables) | average TEDS (%): 81.7 -> 91.2 | Improved |
| use image input instead of PDF input for Acrobat Pro | Acrobat Pro · PubTabNet test (all tables) | average TEDS (%): 65.3 -> 53.7 | Hurt |
| increase encoder feature-map resolution (last CNN layer stride 1) (ablation) | EDD · PubTabNet validation | average TEDS (%) | Improved |
| use independent last CNN layers for structure and cell decoders (ablation) | EDD · PubTabNet validation | average TEDS (%) | Improved |
| EDD instead of TIES on synthetic category C3 | EDD · Synthetic table dataset (category C3) | exact match (%): 52.9 -> 97.2 | Improved |
| EDD instead of TIES on synthetic category C4 | EDD · Synthetic table dataset (category C4) | exact match (%): 68.5 -> 98.0 | Improved |
| EDD instead of TIES on synthetic category C1 | EDD · Synthetic table dataset (category C1) | exact match (%): 96.9 -> 99.7 | Improved |
| shift 90% of first-row cells (TEDS vs adjacency F1 sensitivity) | TEDS metric · PubTabNet validation perturbation | TEDS: 60% drop on TEDS; adjacency F1 nearly 80% | Hurt |
| increase character perturbation level from 10% to 90% | TEDS metric · PubTabNet validation perturbation | TEDS: 90% -> 40% | Hurt |
Closest papers by meaning
- TDeLTA: A Light-Weight and Robust Table Detection Method Based on Learning Text Arrangement (AAAI 2024 · similarity 0.88)
- TRivia: Self-supervised Fine-tuning of Vision-Language Models for Table Recognition (CVPR 2026 · similarity 0.87)
- Grab What You Need: Rethinking Complex Table Structure Recognition with Flexible Components Deliberation (AAAI 2024 · similarity 0.85)
- Multimodal Tabular Reasoning with Privileged Structured Information (NeurIPS 2025 · similarity 0.85)
- TableDART: Dynamic Adaptive Multi-Modal Routing for Table Understanding (ICLR 2026 · similarity 0.84)
- Polynomial-based Self-Attention for Table Representation Learning (ICML 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
