method: Alchera AI2019-05-01

Authors: Hotaek Han, Sewoon Kim, Ja-yeon Jeong

Description: Our method is based on sequence-to-sequence framework.
We use the CNN as a feature extractor in the encoder part and use the attention-based model in the decoder part.
Specifically, We use the focal loss for imbalanced data and use the regional attention for arbitrary shape text.