NVIDIA researchers are back with another project that uses artificial intelligence to generate entirely new images from existing source images. Unlike past work that involved portraits of humans, however, this latest work — called GANimal — transforms an image of an animal into different animals, including other species.
Using an AI technique called generative adversarial networks (GANs), among other things, the researchers developed GANimal, an app that takes the expression of an animal from an image and recreates it on an image of a different animal. Examples include taking an expression from one breed of dog and replicating it on other dog breeds.
Though this is a fun example of the underlying technology, NVIDIA researchers say it could one day be put to use in more serious work. One given example of GANimal’s potential use is engaging filmmakers to shoot images of a tame animal doing stunts, such as a dog, and then using the AI to apply those movements onto a less tame animal, such as a tiger.
NVIDIA’s past artificial intelligence research includes an AI that can accurately scrub noise from images, generate portraits from source images, and transform simple sketches into photorealistic photos.
Author: Go to Source