Neural networks have learned to create a human face. The corresponding project has launched one of the developers of Uber, which employs a combination of two neural networks to create site-generator realistic. And the service is able to create the image of a man and an animal.
The project is based on the algorithm of generative neural networks StyleGAN developed by Nvidia. One of the neural networks responsible for the creation of the image, and the second analyzes the image for realism. When generating a faces the neural network uses the features of several people, and then adapts them to the desired age, facial expressions, hair length, and other parameters.
It’s all fake people
To generate a random person, just go to the website — each time you refresh the page, the neural network will show a different image. We have tested and in some cases is difficult to distinguish from the real thing. The generator first determines the attributes of a human face (the smile to the hair and even freckles), and then connects them together — but so that the image looked like this.
However, sometimes what the image created by the algorithm can be seen immediately: this person, for example, is not realistic.
Also a neural network can be wrong, giving a few individuals in a single frame or combining features of the elderly person with the image of the young man.
In this video you can see how the neural network creates a new entity from the data available.
Although the algorithm operates on a powerful PC with 11 GB of RAM and multiple graphics cards, you can go to thispersondoesnotexist.com and get your own generated image, and with a computer and a smartphone. Did you try it? Write that have happened in our Telegram chat.