Authors: Jianqiang Liu, Hanfei Xu, Bin Zheng, Longhuang Wu, Shangxuan Tian, Pengfei Xiong

Affiliation: Shopee MMU OCR

Description: Our method adopts a transformer-based context-aware framework. We utilize a hybrid architecture encoder and a context-aware autoregressive decoder to construct the recognition pipeline. Finally, a simple but effective multi-model fusion strategy is adopted.

Authors: San San

Description: We use the public synthetic data and the real data. The model is trained separately by CTC and attention, and finally, the multi model fusion strategy is adopted.

Authors: MiaoWang，DanDan Li, Tao Wei，Hongyuan Zhang，FengGu，XuDong Rao

Description: Our method is based on 2D-attention, we simply use ResNet as backbone and a tailored 2D-attention module is applied. The result is generated by a single model without ensemble tricks.