Authors: Wei Wang, Jin Wei, Chengxiang Ran, Xinye Yang, Tianjiao Cao, Fangmin Zhao, Yu Zhou

Affiliation:  Institute of Information Engineering, Chinese Academy of Sciences; Mashang Consumer Finance Co., Ltd

Description: Two Stage method. The detector is a one stage method based on bezier regression and trained with 10k synthtic seal images and train set. The recognizer is TROCR and trained with Chinese recognition benchmark data(https://github.com/FudanVI/benchmarking-chinese-text-recognition) and train set. The synthtic images are generated by taking WTW dataset(https://github.com/wangwen-whu/WTW-Dataset/) as background images and Chinese Company-Names-Corpus(https://github.com/wainshine/Company-Names-Corpus) as corpus.