This repository logs the brief summarization of paper readings for the text style transfer.
This paper deals with the stylish image captioning. The problem is similar to the idea about the stylish news headline summarization. The key to the success of this method relies on the strong assumption that encoded image hidden state and the encoded text hidden state must reside in the same latent space, however, I do not see any mechanism from this paper that can enforce this point. I am also not satisfied by the shown example in this paper, which makes me more doubtful of this method.