method: CUTeOCR2019-04-30

Authors: Zhicheng Yang, Zhuotao Tian, Pengyuan Lyu, Ruiyu Li

Description: We merge predictions of two different alogrithms. One is based on cnn encoder and rnn attention decoder. The other is based on cnn encoder and parallel decoder. We are from HIT and CUHK and Tencent Youtu Lab.