This neural network knows how you will look like in 50 years. How does she work?

It's no secret that now neural networks are most oftenused in photography. We have already seen how they are able to work with animation on a regular computer, and the pictures written by artificial intelligence go under the hammer more expensive than the works of famous artists. Not surprisingly, it was neural networks that formed the basis of the FaceApp application, which in a matter of days gained popularity all over the world, storming the App Store and Google Play stores.

Ilon Musk after Mars colonized

Why the photo? The answer is really simple - there are a lot of pictures and various images on the web. And not just a lot, their billions! Probably, the number of photos has long exceeded the number of the population of the Earth, and even if not all images are laid out on the Internet, there are still a lot of them. Therefore, the neural network is much easier to train on the basis of photos: it can “feed” a lot of data that is in the public domain. This is how the developers of FaceApp managed to achieve impressive accuracy in video processing - the application is very well able to artificially “grow old” and rejuvenate people, change hairstyle, and put on makeup — there are quite a lot of possibilities.

Our compatriot is behind the development of FaceApp- Yaroslav Goncharov, a native of Yandex, who founded the Wireless Lab company. Several projects of the new company were associated with in-depth training of neural networks, and it was FaceApp that “fired”, because it is photos and videos that will become the most important type of information in the near future. With the help of the analysis of patterns in the appearance of people, FaceApp creates new images in which a person looks younger, adult or smiling, for example.

Jeff Bezos after sold to Amazon

The application became popular after a fewmonths after the release in early 2017. FaceApp held on top of the App Store and Google Play in Russia, the USA, the UK, France, Germany and many other countries. Soon, the application reached 40 million downloads. Interestingly, the new round of popularity FaceApp brought exactly the filter "aging". Many young people wondered how they would look after 30, 40 and even 60 years, and FaceApp at the expense of the accuracy of their algorithms allowed them to do it.

Developers do not disclose the algorithm used(unsurprisingly). Perhaps the project is based on the StyleGAN generative neural network algorithm developed by Nvidia. It was he who was involved in the project thispersondoesnotexist, which generated a realistic person who actually never existed. True, in that case, the work of two neural networks took place at once: one of them was directly responsible for creating the image, after which the second analyzed the image for realism. When generating a face, a neural network uses the features of several people at once, and then adapts them with the right age, facial expressions, hair length and other parameters.

In the case of FaceApp, probably the generator firstdefines the attributes of a human face (from smile to hair and even freckles), and then imposes on it the attributes of an old / young face (like wrinkles or, on the contrary, smooth skin) - but so that the image looks like a real one.

26-year-old author of after he decided to meet old age in Switzerland

Read also - Yandex Neural Network has written a play for a symphony orchestra

How to download FaceApp

The application can be downloaded for free for iPhone (here) and Android (here). Initially, the initial set of tools is available, but if you subscribe, get advanced filters.

FaceApp collect user data?

When you use services based onneural networks are lying, they must understand that each “fed” megabyte of data is used for further learning of the neural network. So in a sense, the application does collect user data. And here there are two sides of the medal: on the one, you only want to keep your photos with yourself; on the other hand, the neural network must somehow be taught, because without this, FaceApp could not change photos so realistically.

Of course, for good developers shouldwarn that they are collecting data, otherwise someone might think that they will later be resold and marketers will be used for contextual advertising. Or a photo by the author of will appear in an advertisement of a single bank for pensioners (I hope not).

I wonder how far the neural networks will move inthe next few years. Just two years ago, Lyrebird from Montreal created an AI-based speech synthesizer capable of reproducing any voice. To follow the system, it takes only a few seconds to audio the voice of the required person, on the basis of which the sound fragment will be created. The exact imitation of the voice is possible through the use of neural networks based on artificial intelligence, working on the same principles as the neural networks of the human brain. AI learns to recognize the characteristics of human speech, and then these data are already used to synthesize an artificial voice. So selfies are far from the limit of using neural networks.

But data collection is worth considering.