- Task 1 - E2E Complex Entity Linking
- Task 2 - E2E Complex Entity Labeling
- Task 3 - E2E Zero-shot Structured Text Extraction
- Task 4 - E2E Few-shot Structured Text Extraction
method: Super_KVer2023-03-16
Authors: Lele Xie, Zuming Huang, Boqian Xia, Yu Wang, Yadong Li, Hongbin Wang, Jingdong Chen
Affiliation: Ant Group
Email: yule.xll@antgroup.com
Description: An ensemble of both discriminated and generated models. The former is a multimodal method which utilizes text, layout and image, and we train this model with two different sequence lengths, 2048 and 512 respectively. The texts and boxes are generated by independent OCR models. The latter model is an end-to-end method which directly generates K-V pairs for an input image.
method: Meituan OCR V52025-08-18
Authors: jianqiang liu, boming chen, kai zhou, chen duan, shuaishuai chang, ran wei, shan guo
Affiliation: Meituan
Description: We are Meituan-OCR team. Our method follows the framework of LiLT and designs a jointly training scheme in which SER and RE task are optimized jointly, benefiting greatly for task SER and RE respectively. We only use SVRD-2023 dataset to train the above-mentioned model. In the post-process stage, we design some rules to merge the keys pointing to the same value. Specifically, the table recognition result is served as an alternative information to handle the table format page. Besides, we design an adaptive-inference scheme to tackle the situation in which the relations between long-range keys and values are missing because they are in different 512-token inference items.
method: End-to-end document relationship extraction (single-model)2023-03-15
Authors: Huiyan Wu, Pengfei Li, Can Li, Liang Qiao,
Affiliation: Davar-Lab
Description: Our method realized end-to-end information extraction (single-model) through OCR, NER and RE technologies. Text information extracted by OCR and image information are jointly transmitted to NER to identify key and value entities. RE module extracts entity pair relationships through multi-classification.
Where NER and RE are based on LayoutlmV3, and our training dataset is Hust-Cell.
| Date | Method | Score1 | Score2 | Score | |||
|---|---|---|---|---|---|---|---|
| 2023-03-16 | Super_KVer | 49.93% | 62.97% | 56.45% | |||
| 2025-08-18 | Meituan OCR V5 | 45.28% | 57.50% | 51.39% | |||
| 2023-03-15 | End-to-end document relationship extraction (single-model) | 43.55% | 57.90% | 50.73% | |||
| 2023-03-16 | sample-3 | 42.52% | 56.68% | 49.60% | |||
| 2023-03-16 | sample-1 | 42.13% | 56.36% | 49.25% | |||
| 2023-03-16 | Pre-trained model based fullpipe pair extraction (opti_v3, no inf_aug) | 42.17% | 55.63% | 48.90% | |||
| 2023-03-16 | Pre-trained model based fullpipe pair extraction (opti_v2, no inf_aug) | 42.10% | 55.56% | 48.83% | |||
| 2023-03-16 | Pre-trained model based fullpipe pair extraction (opti_v2, inf_aug) | 42.01% | 55.50% | 48.76% | |||
| 2023-03-15 | Pre-trained model based fullpipe pair extraction (opti_v1) | 41.56% | 55.34% | 48.45% | |||
| 2023-03-16 | Meituan OCR V4 | 41.10% | 54.55% | 47.83% | |||
| 2023-03-16 | Meituan OCR V3 | 40.67% | 54.17% | 47.42% | |||
| 2023-03-15 | Meituan OCR V2 | 40.97% | 53.47% | 47.22% | |||
| 2023-03-16 | submit-trainall | 40.65% | 52.98% | 46.82% | |||
| 2023-03-16 | submit-ocrkie-8to2 | 40.15% | 52.97% | 46.56% | |||
| 2023-03-14 | Meituan OCR | 39.85% | 52.46% | 46.15% | |||
| 2023-03-16 | f2 | 41.07% | 50.82% | 45.94% | |||
| 2023-03-16 | final | 41.05% | 50.80% | 45.93% | |||
| 2023-03-16 | submit-8finetune2 | 39.58% | 51.93% | 45.75% | |||
| 2023-03-15 | new-model | 39.38% | 48.59% | 43.99% | |||
| 2023-03-15 | 800-fix2 | 37.06% | 46.46% | 41.76% | |||
| 2023-03-11 | add-pplssm | 36.45% | 43.83% | 40.14% | |||
| 2023-03-16 | LayoutLM & STrucText Based Method | 33.09% | 45.92% | 39.51% | |||
| 2023-03-15 | bug-800 | 34.17% | 43.91% | 39.04% | |||
| 2023-03-16 | Layoutlmv3 | 29.81% | 41.45% | 35.63% | |||
| 2023-03-15 | old-500-fix1 | 27.64% | 35.52% | 31.58% | |||
| 2023-03-15 | 数据之关联2 | 23.26% | 35.07% | 29.16% | |||
| 2023-03-16 | 处理t | 17.34% | 26.92% | 22.13% | |||
| 2023-03-16 | refinet | 17.11% | 26.60% | 21.86% | |||
| 2023-03-12 | FirstResult | 16.51% | 26.12% | 21.32% | |||
| 2023-03-16 | 不处理t的结果 | 16.39% | 25.56% | 20.97% | |||
| 2023-03-15 | 表格结构分析+layout的结果_0315 | 16.25% | 25.38% | 20.81% | |||
| 2023-03-15 | 数据之关联 | 16.75% | 24.48% | 20.61% | |||
| 2023-03-16 | Ant-FinCV | 14.44% | 22.68% | 18.56% | |||
| 2023-03-16 | Ant-FinCV | 14.32% | 22.70% | 18.51% | |||
| 2023-03-16 | Ant-FinCV | 14.38% | 22.62% | 18.50% | |||
| 2023-03-16 | Ant-FinCV | 14.21% | 22.35% | 18.28% | |||
| 2023-03-16 | Ant-FinCV | 13.79% | 21.75% | 17.77% | |||
| 2023-03-15 | layoutxlm-relation and ppstructure box level | 12.86% | 21.56% | 17.21% | |||
| 2023-03-15 | vocr | 11.71% | 19.13% | 15.42% | |||
| 2023-03-13 | FIne tuned DONUT | 13.06% | 17.15% | 15.11% | |||
| 2023-03-14 | Layoutlm relation extraction | 10.99% | 19.22% | 15.10% | |||
| 2023-03-14 | layoutxlm and ppstructure | 11.63% | 18.43% | 15.03% | |||
| 2023-03-15 | layoutxlm-relation and ppstructure token level | 11.51% | 18.26% | 14.89% | |||
| 2023-03-14 | vocr | 10.31% | 17.53% | 13.92% | |||
| 2023-03-16 | Ant-FinCV | 8.96% | 14.84% | 11.90% | |||
| 2023-03-14 | e2e | 1.77% | 3.44% | 2.60% | |||
| 2023-03-13 | first commit | 1.22% | 2.33% | 1.78% | |||
| 2023-03-14 | e2e | 0.55% | 1.01% | 0.78% | |||
| 2023-03-10 | test | 0.00% | 0.00% | 0.00% | |||
| 2023-03-11 | test_t1 | 0.00% | 0.00% | 0.00% | |||
| 2023-03-13 | intime | 0.00% | 0.00% | 0.00% | |||
| 2023-03-13 | test2 | 0.00% | 0.00% | 0.00% | |||
| 2023-09-14 | Graph Attention | 0.00% | 0.00% | 0.00% |