Neural networks programming draws pictures
It seems that not a week goes by when the concepts of artificial intelligence and neural networks do not sprinkle in the headlines of the media. Usually they cover a new exciting use case, or in the article the author discusses whether the impressive achievements of AI will replace human labor. And here the neural network reached human art.
It is difficult to become a real artist – not only certain knowledge and skills, but also talent are needed. That is why not everyone can become an artist, and much less the machine. But with more modernized version of the arts, photoshop, everything is easier. The main task is to understand the intricacies of buttons and windows. And here not only the amateur artist will deal with it, but also artificial intelligence will understand the familiar line of code. It was with this idea that scientists from the Massachusetts Institute of Technology, as well as researchers from IBM Watson AI Lab took up the project of Generative Advertising Networks.
Neural network programming is now used in art
What is the essence of the project: the user has the original picture, certain items and actions that he can apply to them. For example, the user has a photo of a night street and a lantern. He chooses a lantern, highlights the area where he wants to add it and that’s all. The rest will be done by the neural network GAN.
According to those who already had a chance to use the new system, the innovation has some nuances. It is doing great to remove or add objects of wildlife, for example, the foliage or the sky. But more difficult with the buildings – geometric shapes are distorted and the facade looks clumsy. But still the fact remains – thanks to neural networks, everyone now has a chance to draw pretty good pictures.
Experts also remain positive about the technology. These are only initial steps of the neural network in this direction. Besides, as it is known, the first computers were not perfect either. A few years ago, the computer took several rooms, and now fits in your pocket. According to users, this is just an initial presentation of the technology’s capabilities. If it is developed, it will still be able to surprise the world, experts say.
By the way, anyone can try to draw with the neural network. On the website of the developers there is an online version of this innovation.
It is also worth noting that this is not the first such project. Prior to that, Google AutoDraw also released a neural network platform aimed at drawing. However, there the neural network analyzes the strokes, and then redraws them according to the original idea of the author. Thus, AutoDraw is able to turn doodles into full-fledged drawings.
Google-based AI editor and machine learning guesses what the person wanted to draw (for example, a dog, a house, or a ship), and “draws” his idea in a more presentable way. So, the algorithms are smart enough to guess to make a fish out of two triangles.