Neural network creates paintings in the style of Van Gogh and Picasso



    Google recently showed how a neural network can independently create works of art. Simply put, paint pictures by exaggerating existing images.

    A new experiment conducted at the University of Tübingen (Germany) demonstrates an alternative algorithm for the neural network: it plausibly fakes the art style of Vincent Van Gogh, Pablo Picasso, Edward Munch and any other artists. Any images are suitable for processing input.

    An example of the operation of the algorithm is shown in the illustrations above, where the neural network processed in different styles a photograph of a street in a German town.

    The generated paintings are really beautiful and, without a doubt, have 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. On the contrary, the value of the scripts will only increase.

    The neural network has 19 nested layers, and the processing of the original image occurs in several stages. At each stage in the hierarchy, the number of filters increases, but at the same time the image size becomes smaller due to the implementation of a kind of down-sampling mechanism.



    Styling for a specific author occurs at stages d , e .

    The authors of the scientific work consider the main achievement that it was possible to successfully separate the representations of content and style in the neural network. Thus, they can be controlled independently of each other. For example, take content from one image, and style from another. This is what is shown in the very first illustration, where one photograph is converted into paintings made in different styles.

    Layers with style are applied either to the whole image, or fragmented. Here's what a photograph of the city looks like if you apply the style of the famous painting “Composition VII” by Vasily Kandinsky to it.

    Original


    Neural Network Result

    Also popular now: