Authors: Wenqi Zhao

Affiliation: Wangxuan Institute of Computer Technology, Peking University, Beijing, China

Email: 1027572886a@gmail.com

Description: We only use annotation and OCR results. Build our model using simpletransformer package.
In the data preprocess phrase, we use fuzzy search to find the start index of the answer span.
Refer to:
https://simpletransformers.ai/
https://github.com/mineshmathew/DocVQA/tree/master/BERT_baseline
https://github.com/taleinat/fuzzysearch