method: Double-U 2022-07-18

Authors: Daehee Kim, Yumin Lim, Yoonsik Kim

Description: Our proposed method, Double-U, consists of an transformer based encoder-decoder structure. Our method can train and infer document understanding tasks including localization in the same architecture. Double-U provides multiple models as encoders and decoders, but we used swin transformer as an encoder and BERT as a decoder for this competition.