method: TWA2022-03-15

Authors: Zan-Xia Jin, Mike Zheng Shou, Fang Zhou, Satoshi Tsutsui, Jingyan Qin, Xu-Cheng Yin

Affiliation: University of Science and Technology Beijingļ¼Œ National University of Singapore

Description: We propose an OCR Token-Word Contrastive (TWC) learning task, which pre-trains word representation by augmenting OCR tokens via the Levenshtein distance between the OCR tokens and words in a dictionary.

Jin Z X, Shou M Z, Zhou F, et al. From Token to Word: OCR Token Evolution via Contrastive Learning and Semantic Matching for Text-VQA[C]//Proceedings of the 30th ACM International Conference on Multimedia. 2022: 4564-4572.