method: LayoutLMv32023-03-13

Authors: sofiawu, morajiang

Affiliation: TencentOCR

Description: Based on a large pretrained model and LayoutLM v3 architecture, with some pre/post processing methods.

method: KIE-Brain32023-03-17

Authors: Boqian Xia, Yu Wang, Yadong Li,Ruyi Zhao, Zuming Huang, Lele Xie, Jingdong Chen, Hongbin Wang

Affiliation: Ant Group

Email: xiaboqian.xbq@antgroup.com, yuwangtj@yeah.net

Description: An ensemble of multi-task end-to-end information extraction models. The document question answering task and the document information extraction task are jointly realized, and the model performance is improved. At the same time, this solution is an end-to-end information extraction method and does not rely on external OCR.

method: KIE-Brainer22023-03-17

Authors: Boqian Xia, Yu Wang, Yadong Li,Ruyi Zhao, Zuming Huang, Lele Xie, Jingdong Chen, Hongbin Wang

Affiliation: Ant Group

Email: xiaboqian.xbq@antgroup.com, yuwangtj@yeah.net

Description: An ensemble of multi-task end-to-end information extraction models. The document question answering task and the document information extraction task are jointly realized, and the model performance is improved. At the same time, this solution is an end-to-end information extraction method and does not rely on external OCR.

Ranking Table

Description Paper Source Code
DateMethodscorescore1score2
2023-03-13LayoutLMv376.90%79.58%66.20%
2023-03-17KIE-Brain371.44%74.90%57.59%
2023-03-17KIE-Brainer271.24%74.82%56.92%
2023-03-17KIE-Brain71.24%74.87%56.69%
2023-03-13Donut_VIE1.37%1.47%1.01%

Ranking Graphic