Deep Learning Looks To Take Over Artist Jobs Now

Artificial Intelligence  

Artificial intelligence researchers have created an deep neural network system that creates artistic images of that look like they could have been created by famous artists. The system uses neural representations to separate and recombine content and style, in the likes of Picasso, Munch, Kandinsky and others.

Researchers at the University of Tubingen in Germany have created an artificial intelligence system based on a deep neural network that creates artistic images of high perceptual quality. The system uses neural representations to separate and recombine content and style of arbitrary images, providing a neural algorithm for the creation of artistic images.

As the images above and below demonstrate, the algorithm can apply a very smart filter on a photograph so that it duplicates the style of a Van Gogh painting, or uses the brush strokes and colors of Edvard Munch's The Scream, or any number of other famous artistic style.

According to the computational neuroscience researchers, Leon A. Gatys, Alexander S. Ecker, and Matthias Bethge, responsible for the study, the striking similarities between their performance-optimized artificial neural networks and traditional human-made artwork "offers a path forward to an algorithmic understanding of how humans create and perceive artistic imagery."

In these examples the initial input photograph (top left) is reproduced in Pablo Picasso's Cubist style, and Wassily Kandinsky's.

To obtain a representation of the style of an input image, the researchers used a feature space originally designed to capture texture information. This feature space is built on top of the filter responses in each layer of the network. It consists of the correlations between the different filter responses over the spatial extent of the feature maps.

Related articles The Rise of Artificial Intelligence  James Miller Discusses Aspects of the Singularity  Stuart Armstrong and Nick Hawes Discuss the Future of Artificial Intelligence

By including the feature correlations of multiple layers, they were able to create a stationary, multi-scale representation of the input image, capturing its texture information but not the global arrangement.

"In light of the striking similarities between performance-optimised artificial neural networks and biological vision, our work offers a path forward to an algorithmic understanding of how humans create and perceive artistic imagery."

It is truly fascinating that a convolutional neural network, which was trained to perform one of the core computational tasks of biological vision, automatically learns image representations that allow the separation of image content from style.

The  researchers suggest anexplanation could be that when learning object recognition, the network has to become invariant to all image variation that preserves object identity. Representations that mathematically represent these variations in the content of an image and the variation in its appearance would be extremely practical for this task.

"Thus, our ability to abstract content from style and therefore our ability to create and enjoy art might be primarily a preeminent signature of the powerful inference capabilities of our visual system."

According to one poster on a reddit thread about the research, "Art, among the few jobs that people thought AI would take the longest to replace, looks like it will be among the first to be replaced by AI."

At the very least, Photoshop filters of the not-to-distant future should be pretty incredible.

SOURCE  arXiv.org

By 33rd Square

Embed