method: BreSee OCR2021-07-06

Authors: Tianpeng Li, Linzhi Zhuang, Mengyue Shao, Jie Wu, Jiling Wu

Affiliation: BreSee AI Lab, Zhejiang Sci-Tech University

Description: We modify the SRN network structure and loss function, synthesize a large number of data sets, train SRN and fine tune on the training set, and use a large number of data enhancement operations.

method: H&H Lab2019-04-22

Authors: HUST_VLRGROUP(Hui Zhang, Mingkun Yang, Mengde Xu, Zhen Zhu, Jiehua Yang) & HUAWEI_CLOUD_EI(Jing Wang, Yibin Ye, Shenggao Zhu, Dandan Tu)

Description: We mainly completed the task 2 using CRNN. Different from the content in the paper, we modified the structure of CNN to PVANet-like and used multiple GRU layers. And we adjusted the training strategy to continue to improve the recognition result.

method: CLOVA OCR2019-04-22

Authors: Sungrae Park, Seung Shin, Seonghyeon Kim, Jaeheung Surh, Junyeop Lee, Hwalsuk Lee

Description: Our model consists of a ResNet-based backbone, a sequence model, and an attention-based decoder [1]. The backbone is a combination of the ResNet and SENet(squeeze and excitation network) [2] and the others are based on Baek et al. [1]. We trained the model with our own synthetic datasets by applying virtual adversarial training (VAT) techniques [3]. For this competition, we fine-tuned the model with the training dataset of SROIE. The recognition identified the texts on the detected text boxes by CRAFTS [4].

Ranking Table

Description Paper Source Code
2021-07-06BreSee OCR96.78%96.92%96.85%
2019-04-22H&H Lab96.35%96.52%96.43%
2019-04-22CLOVA OCR94.30%94.88%94.59%
2019-04-22A Text Extraction Method Based on Modified CRNN26.33%72.53%38.63%

Ranking Graphic