Neural network creates a picture in the style of Van Gogh and Picasso

Google has recently showed , as the neural network can create their own works of art. Simply put, draw pictures, exaggerating existing images.

new experiment , conducted by the University of Tübingen (Germany) shows an alternative algorithm for neural network: it is plausible forges artistic style of Vincent Van Gogh, Pablo Picasso, Edvard Munch and any other artists. The input processing suitable for any images.

An example of the algorithm is an illustration at the top, where the neural network is processed in different styles photo street in the German town.

Generated pictures are really beautiful, no doubt, have a high artistic value. However, this does not mean that the originals of the works of the great masters of the past will fall in price. Rather, the value of the originals only grow.

The neural network has 19 sub-layers, and the original image processing takes place in several stages. At each stage in the hierarchy of increasing the number of filters, but the image size becomes smaller due to the implementation of the mechanism of a sort of downsampling.



Styling for a particular author at a step d i>, e i>.

Authors of scientific work is considered a major achievement that successfully separate the representation of content and style in the neural network. Thus, they can be controlled independently. For example, take content from one image and style - from another. It is shown in the very first illustration, where one picture is converted into a pattern made in different styles.

The layers are applied with a style, or to the entire image, or in fragments. Here is a picture of the city, if you apply a style to her famous painting "Composition VII» Wassily Kandinsky.

Original


The output of the neural network




Source: geektimes.ru/post/261116/