Since somthing went wrong running takase's code under the circumstances of pytorch 1.4.0 (which my gpu doesn't supports T.T ). I've reimplemented it on the newest fairseq
实现了论文 https://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Gatys_Image_Style_Transfer_CVPR_2016_paper.pdf 中的方法