This repository provides a curated list of resources in the research domain of Table Structure Recognition (TSR), updated in my free time.
Methods based on heuristic rules is not included.
"Table Structure Recognition" OR "Table Recognition"
- Methodology
- Review
- Top-Down
- Bottom-Up
- Cell based / Grid based
- Word based / Text-Line based
- Image-to-Sequence / End-to-End
- Data
- Metric
- Dataset / Benchmark
- Data Representation
- Data Synthesis / Data Generation
- Data Augmentation
- Supervised training only using the training set of PubTabNet Version 2 (500777 samples)
- Evaluation for PubTabNet:
- High TEDS
- High TEDS-S + High AP: to focus on the structure
- High TEDS-S only: insufficient
- It is allowed to use additional third-party data or pre-trained models for performance improvement.
- HTML tags that define the text style including bold, italic, strike through, superscript, and subscript should be included in the cell content.
- Due to a problem with the final evaluation data set, bold tags
<b>where not considered in the evaluation.
| Paper | Group | Date | Publication | TEDS | TEDS-S | AP-50 |
|---|---|---|---|---|---|---|
TFLOP TFLOP: Table Structure Recognition Framework with Layout Pointer Mechanism |
Upstage AI | 2024-08 | IJCAI-2024 | 96.66 | 98.38 | - |
MuTabNet Multi-Cell Decoder and Mutual Learning for Table Structure and Character Recognition |
PFN | 2024-04 | ICDAR-2024 | 96.53, Simple-98.01, Complex-94.98 | - | - |
| An End-to-End Local Attention Based Model for Table Recognition | NII | 2023-07 | ICDAR-2023 | 96.21, Simple-97.77, Complex-94.58 | - | - |
MTL-TabNet An End-to-End Multi-Task Learning Model for Image-based Table Recognition |
NII | 2023-03 | VISIGRAPP-2023 | 96.17, Simple-97.60, Complex-94.68 | - | - |
WSTabNet Rethinking Image-based Table Recognition Using Weakly Supervised Methods |
NII | 2023-02 | ICPRAM-2023 | 95.97, Simple-97.51, Complex-94.37 | - | - |
CoT_SRN Contextual transformer sequence-based recognition network for medical examination reports |
SDNU | 2022-12 | Applied-Intelligence.2023 | 92.34 | 95.71 | - |
| Team | Group | TEDS |
|---|---|---|
| Davar-Lab-OCR | Hikvision | 96.36, Simple-97.88, Complex-94.78 |
| VCGroup | Ping An | 96.32, Simple-97.90, Complex-94.68 |
| XM | USTC-NELSLIP | 96.27, Simple-97.60, Complex-94.89 |
| YG | 96.11, Simple-97.38, Complex-94.79 | |
| DBJ | 95.66, Simple-97.39, Complex-93.87 | |
| TAL | TAL | 95.65, Simple-97.30, Complex-93.93 |
| PaodingAI | Paoding | 95.61, Simple-97.35, Complex-93.79 |
| anyone | 95.23, Simple-96.95, Complex-93.43 | |
| LTIAYN | 94.84, Simple-97.18, Complex-92.40 |
- Supervised training only using the training set of FinTabNet Version 1.0.0 (91596 samples)
- Evaluation for PubTabNet:
- High TEDS
- High TEDS-S only: insufficient