Research map / Vision-language and multimodal / Human and embodied modeling
TDATR: Improving End-to-End Table Recognition via Table Detail-Aware Learning and Cell-Level Visual Alignment
Chunxia Qin, Chenyu Liu, Pengcheng Xia, Jun Du, Baocai Yin, Bing Yin, Cong Liu
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 |
|---|---|---|---|
| add table content recognition tasks to detail-aware learning (ablation) | TDATR · iFLYTAB-full | TEDS: 82.63 -> 91.57 | Improved |
| add table structure understanding tasks (ablation) | TDATR · iFLYTAB-full | TEDS: 82.63 -> 90.44 | Improved |
| combine content and structure tasks (full two-stage) (ablation) | TDATR · PubTabNet | TEDS: 89.19 -> 94.38 | Improved |
| add content tasks on top of structure tasks (ablation) | TDATR · iFLYTAB-full | TEDS: 90.44 -> 92.50 | Improved |
| add structure-guided refinement with multi-resolution features (ablation) | SGCL · iFLYTAB-full | AP50: 89.22 -> 94.37 | Improved |
| add structure mask to bidirectional enhancement (ablation) | SGCL · PubTabNet | AP50: 90.02 -> 90.26 | Improved |
| add structure mask to bidirectional enhancement (ablation) | SGCL · iFLYTAB-full | TEDS: 93.51 -> 93.41 | No clear effect |
| add cell localization module (C4) to HTML parsing (ablation) | TDATR · PubTabNet | TEDS: 94.38 -> 94.80 | Improved |
| use TDATR instead of Dolphin (same modeling approach) | end-to-end TR · PubTables-1M | TEDS: 95.48 -> 97.97 | Improved |
| use TDATR instead of TableFormer (24x more data) | TSR · PubTabNet-Val | TEDS-S: 96.75 vs 96.27 | Hurt |
| fine-tune two more epochs on PubTabNet (TDATR-ft) (ablation) | TDATR · PubTabNet-Val | TEDS: 95.12 -> 96.10 | Improved |
| use TDATR instead of PaddleOCR-VL | table recognition · OmniDocBench 1.5 | TEDS: 91.95 vs 87.96 | Hurt |
Closest papers by meaning
- TDeLTA: A Light-Weight and Robust Table Detection Method Based on Learning Text Arrangement (AAAI 2024 · similarity 0.89)
- TRivia: Self-supervised Fine-tuning of Vision-Language Models for Table Recognition (CVPR 2026 · similarity 0.88)
- TabPedia: Towards Comprehensive Visual Table Understanding with Concept Synergy (NeurIPS 2024 · similarity 0.87)
- TableDART: Dynamic Adaptive Multi-Modal Routing for Table Understanding (ICLR 2026 · similarity 0.87)
- TABLET: A Large-Scale Dataset for Robust Visual Table Understanding (ICLR 2026 · similarity 0.87)
- NeedleInATable: Exploring Long-Context Capability of Large Language Models towards Long-Structured Tables (NeurIPS 2025 · similarity 0.86)
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.
