In October 2018, the Google division underDeepMind created the BigGAN neural network for creating photos. NVIDIA went further, and used a similar algorithm to create the faces of people who do not exist. The result of the neural network is shown in the video - it is both impressive and scary. The fact is that computer-generated portraits are no different from real ones, and there is hardly a person who can immediately recognize a fake.
As in many similar projects, developersused generative-adversary neural network GAN. One part of it acts as an image generator, and the second assesses their similarity with samples of the ideal result. CelebA-HQ celebrity photos were used as samples. It consists of about 30,000 images, so there was clearly no shortage of training facilities for developers.
To achieve "unprecedented quality",Neural network first created images in low resolution. As the results of his work improved, programmers added new layers with higher resolution. The process of creating images can be viewed in the video below, and read the details in the technical document in PDF format.
The development of artificial intelligence franklyscary. NVIDIA's neural network is far from being the only system that imitates something human. In May 2017, the DeepMind WaveNet project was presented, ideally replicating human speech. Also, developers are actively developing robots that can understand and express human emotions. A striking example is the Sophia robot, developed by Hanston Robotics and obtained Saudi Arabian citizenship.
What do you think artificial intelligence is capable of? Write your bold assumptions in the comments or in our Telegram-chat.