what‘s your dataset for training？

[chenj02]

Good work! It seems we need a style image, a text and a same style image (content like the text) as a training set, but it's unclear in the paper? Can you explain more clear? Thanks!