< BackHome/news/instagan-image-to-image-translation

Jean Genie

generative adversarial networks
Swapping a face or object into pictures is nothing new for machines, but when that object to be swapped is a bit more varied—say, a skirt in place of jeans in a fashion spread—it usually still needs to be done by humans to finesse areas such as exposed skin where pant legs used to be. Now “InstaGAN: Instance Aware Image-to-Image Translation,” which uses generative adversarial networks to swap a panoply of different objects into a panoply of different scenarios and images, may automate this task into the stratosphere. Inevitable nefarious propagandistic uses aside, this is a promising development: Imagine instantly trying on a virtual outfit over a series of different images and scenarios in your photo album, and you get the idea of how this might be capitalized upon for online retail some day.
View Full Article >

sign up for our newsletter