About the transformer block

[hello-trouble]

Thank you for your excellent job about the style-transfer. I am a little puzzled about the transformer block of this paper which is different from the code. Why doesn't adopt the fully convolutional neural network?