Nvidia researchers have created an artificial intelligence capable of generating realistic portraits of imaginary individuals through a generative antagonist network (GAN). Such systems have been generating human faces for a while now, but never with this kind of result. The researchers tested the same model on a slew of cat photos, but strangely, kitties seem more resistant to random generation. The researchers note in their paper: “Cats continue to be a difficult dataset due to the high intrinsic variation in poses, zoom levels, and backgrounds.” In short, it takes a certain spatial uniformity for the GAN to shine.
⇨ PetaPixel, “These portraits were made by AI: none of these people exist.”