Making fake hot people with GAN’s

Take a look at the beautiful people below.

No, these are not real celebrities. Much like the beauty standards that our society dictates, these people are completely FAKE.

The chipmaking company NVIDIA created these images of fake celebs using something called a General Adversarial Network or GAN.

Introduced in 2014 by the scientist Ian Goodfellow, a GAN is a type of artificial intelligence algorithm that uses two separate neural networks. A neural network is modeled on the human nervous system. These systems can learn and continually improve their performance over time.

A neural network can learn to recognize photos of turtles, for example, by analyzing tens of thousands of turtle pics. This is great, but before the networks can begin learning, humans need to have carefully labeled what’s pictured in each of these images. This is very time-consuming and very costly, and a lot of turtle pics.

What’s special about the GAN is the second network. One network can help to teach the other, which means less work for humans! With a GAN, one network of the GAN generates the imagery based on the data it’s fed, and the second network — the adversary — checks if they’re real.

For NVIDIA’s project, they fed their first network a database of actual celebrity pictures, which in turn generated the fake images. As you can guess, the second network checks the images for believability. Then the first network improves upon their fakes to make them more believable. The two work in tandem to continually produce more realistic faces.

For Nvidia’s purposes, this technology will most likely be used to create game characters. This GAN could also be a great tool used for advertising. Instead of hiring models or actors for their content, advertisers could just generate their own fake attractive people. With this in mind, it’s not hard to imagine harnessing the power of GAN’s to create convincing fake images, videos, and people in a more sinister way. We’re talking anything from creating fake news “evidence” to running a light catfishing scam.

You can check out NVIDIA’s full paper here. http://research.nvidia.com/publication/2017-10_Progressive-Growing-of