Amazing AI Automatically Turns You Into an Anime Character
And now a team of South Korean researchers from video game company NCSoft has done just that. They created a generative adversarial networks that will ingest your selfies and spit out a picture-perfect cartoon caricature straight out of your favorite anime.
Their new method, called “U-GAT-IT,” bridges the substantial gap between a selfie photograph and a cartoon image by using a framework called “Adaptive Layer-Instance Normalization.” The procedure is able to focus on important qualities of an image like eyes and ears without being distracted by other traits.
And that's a big step forward.
“Unlike previous attention-based methods which cannot handle the geometric changes between domains, our model can translate both images requiring holistic changes and images requiring large shape changes,” reads the researchers' paper, which was submitted to preprint repository arXiv last month.