Human beings are quite smart when it comes to understanding the relationship between different domains. For example, we can easily understand the relationship between a Spanish sentence and its translated version in English. We can even guess which color tie to wear to match a certain kind of suit. While it seems easy for humans, this is not a straightforward process for machines.
The task of style transfer across different domains for machines can be framed as a conditional image generation problem. Given an image from one domain, can we learn to map to an image from a different domain.
While there have been many approaches to achieve this using pairwise labeled data from two different domains, these approaches are fraught with problems. The major issue with these approaches is obtaining the pairwise labeled...
