German researchers have published a paper outlining a new method of using deep learning to turn regular photographs into images that evoke the style of great painters such as Vincent Van Gogh, Edvard Munch, and Picasso.
The researchers' method involves feeding mundane images into a convolution neural network that creates a new image in the style of a given painter. The researchers, led by the University of Tuebingen's Leon Gatys, say the key insight is neural networks can separate the style and the content of a given image, which enables them to extract the "style" of a given painter by analyzing their various works and then apply that style to images with different content.
"The key finding of this paper is that the representations of content and style in the convolutional neural network are separable," the researchers say. "That is, we can manipulate both representations independently to produce new, perceptually meaningful images."
They say their algorithm can produce new images in only an hour and they expect refinements will be able to bring that time down further.
From The Washington Post
View Full Article - May Require Free Registration
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found