This year alone has seen unprecedented leaps in the area of learning-based
image translation, namely cyclegan, by Zhu et al. But experiments so far have
been tailored to merely two domains at a time, and scaling them to more would
require an quadratic number of models to be trained. An