Authors: Meichao Luo, Zhenming Wang, Tao Zhang, Qiang Wang, Xiaogang Huang
Description: We proposed a method based on RARE and ensemble it with a modified CRNN model. Center Loss with initial value was also applied. Besides model modification, we use a huge public dataset for pretrain and generated lots of data with the train set image style for finetune. Finally, we use a language model to correct errors.