method: Applica.ai Lambert 2.0 + Excluding OCR Errors + Fixing total entity2021-01-02
Authors: Applica.ai research team
Affiliation: Applica.ai
Description: Following the same evaluation rules as others, the OCR mismatch errors are excluded in the submission.
Additionally, we have manually fixed annotation discrepancies in "total" entity in the test set.
Note:
1. We submitted the best solution out of 100 fine-tuned models
2. In this task there is an annotation discrepancy in "total" entity which caused unfair comparison between models (In train/test sets "total" entity was randomly prefixed by "RM"). Number of errors in the top solutions caused by this kind of annotation error:
Applica.ai Lambert 2.0 + Excluding OCR Errors + Fixing total entity = 0
LayoutLM 2.0 (single model) = 3 (example: 275)
Applica.ai Lambert 2.0 + Excluding OCR Mismatch = 8 (example: 77)
Tencent Youtu = 8 (example: 120)
VIE = 0
HIK_OCR_Exclude_ocr_mismatch = 0
LayoutLM + Excluding OCR Mismatch = 9 (example: 121)
method: LayoutLM 2.0 (single model)2020-12-24
Authors: LayoutLM Team
Affiliation: LayoutLM Team
Description: Multi-modal Pre-training for Visually-Rich Document Understanding
method: Applica.ai Lambert 2.0 + Excluding OCR Mismatch2021-01-01
Authors: Applica.ai research team
Affiliation: Applica.ai
Description: Upgraded Lambert model (based on RoBERTa-base) trained longer on bigger datasets with extra layout embeddings (sinusoidal + relative) for each subtoken (paper is under preparation).
Following the same evaluation rules as others, the OCR mismatch errors are excluded in the submission.
Note:
1. We submitted the best solution out of 100 fine-tuned models
Date | Method | Recall | Precision | Hmean | |||
---|---|---|---|---|---|---|---|
2021-01-02 | Applica.ai Lambert 2.0 + Excluding OCR Errors + Fixing total entity | 96.83% | 99.56% | 98.17% | |||
2020-12-24 | LayoutLM 2.0 (single model) | 96.61% | 99.04% | 97.81% | |||
2021-01-01 | Applica.ai Lambert 2.0 + Excluding OCR Mismatch | 96.40% | 99.11% | 97.74% | |||
2020-12-07 | Tencent Youtu | 96.47% | 98.89% | 97.67% | |||
2020-12-28 | VIE | 96.33% | 98.53% | 97.41% | |||
2020-05-07 | HIK_OCR_Exclude_ocr_mismatch | 96.33% | 98.38% | 97.34% | |||
2020-04-18 | LayoutLM + Excluding OCR Mismatch | 96.04% | 98.16% | 97.09% | |||
2020-11-09 | admintest | 96.33% | 96.33% | 96.33% | |||
2020-04-15 | PICK-PAPCIC & XZMU | 95.46% | 96.79% | 96.12% | |||
2020-04-16 | LayoutLM | 96.04% | 96.04% | 96.04% | |||
2020-03-26 | Applica.ai roberta-base-2D | 95.39% | 95.80% | 95.60% | |||
2020-06-05 | great | 94.24% | 94.24% | 94.24% | |||
2019-08-14 | PATech_AICenter | 94.02% | 94.02% | 94.02% | |||
2020-05-23 | GIE | 91.21% | 93.43% | 92.31% | |||
2020-07-07 | Taikang Insurance Group Research Institute | 91.79% | 91.99% | 91.89% | |||
2019-08-05 | PATECH_CHENGDU_OCR_V2 | 91.21% | 91.21% | 91.21% | |||
2020-02-20 | Character & Word BiLSTM Encoder | 90.85% | 90.85% | 90.85% | |||
2019-05-05 | Ping An Property & Casualty Insurance Company | 90.49% | 90.49% | 90.49% | |||
2019-04-29 | Enetity detection | 89.70% | 89.70% | 89.70% | |||
2019-05-04 | H&H Lab | 89.63% | 89.63% | 89.63% | |||
2019-05-02 | CLOVA OCR | 89.05% | 89.05% | 89.05% | |||
2020-12-29 | coldog | 86.17% | 86.17% | 86.17% | |||
2019-09-23 | ASTRI-CCT-MSA | 85.45% | 85.45% | 85.45% | |||
2019-05-05 | GraphLayout | 85.09% | 85.09% | 85.09% | |||
2020-06-15 | End-to-end learning with PGN | 83.86% | 83.86% | 83.86% | |||
2019-05-04 | HeReceipt-withoutRM | 83.00% | 83.24% | 83.12% | |||
2020-06-17 | Graph Neural Net with Bert Embeddings | 82.78% | 82.78% | 82.78% | |||
2019-05-06 | BOE_IOT_AIBD | 82.71% | 82.71% | 82.71% | |||
2019-05-05 | PATECH_CHENGDU_OCR | 81.70% | 82.29% | 82.00% | |||
2020-05-28 | SROIE LSTM - Axel Alejandro Ramos GarcĂa | 81.99% | 81.99% | 81.99% | |||
2020-04-28 | BERT-MRC | 81.05% | 81.05% | 81.05% | |||
2020-05-29 | Cool Method Remix | 79.03% | 79.03% | 79.03% | |||
2019-04-30 | NER with spaCy model | 78.96% | 79.02% | 78.99% | |||
2020-12-28 | Custom Named Entity Recognition | 77.59% | 77.59% | 77.59% | |||
2021-01-02 | lstm deep | 77.38% | 77.38% | 77.38% | |||
2019-05-05 | CITlab Argus Information Extraction (positional & line features, enhanced gt) | 77.38% | 77.38% | 77.38% | |||
2021-01-02 | lstm standard method trained 100 epochs constant learning rate | 76.15% | 76.15% | 76.15% | |||
2019-04-28 | A Simple Method for Key Information Extraction as Character-wise Classification with LSTM | 75.58% | 75.58% | 75.58% | |||
2019-04-30 | Bi-directional LSTM-CNNs-CRF (version2) | 74.86% | 74.86% | 74.86% | |||
2019-05-05 | Location-aware BERT model for Text Information Extraction | 74.42% | 74.42% | 74.42% | |||
2020-05-23 | test | 73.63% | 73.63% | 73.63% | |||
2019-04-30 | BERT with Multi-task Confidence Prediction | 66.14% | 66.14% | 66.14% | |||
2019-05-02 | With receipt framing | 63.04% | 63.54% | 63.29% | |||
2019-05-05 | IFLYTEK-textNLP_v2 | 61.24% | 61.24% | 61.24% | |||
2019-05-05 | SituTech_OCR | 59.01% | 62.38% | 60.64% | |||
2019-04-30 | Key Information Extraction from Scanned Receipts | 28.75% | 36.31% | 32.09% |