Computer Vision News - April 2018

and a style image of Obama, the visual attributes will be transferred from Obama to Trump in a semantic manner. One will be able to see in the paper many such examples, in which the shape and color of eyes, eyebrows, skin and lips are transferred. The authors also showed two other applications : domain transfer and puppet control; both are very interesting too. It is nice to note that this work is not using Generative Adversarial Networks (GANs): besides being so popular these days, they also present a lot of problems which need to be overcome to make them work. “ GANs are nice - says one of the authors - but they make the training process much harder! ” Research 31 Research Computer Vision News “GANs are nice, but they make the training process much harder!” We are indebted to the authors Roey Mechrez , Itamar Talmi and Lihi Zelnik- Manor for allowing us to use images from their paper to illustrate this article. “How to make it derivable, so that it can be used as a loss function”