Neural Style Transfer

Neural style transfer is a class of software algorithms that use deep neural networks to manipulate images and video, combining the content of one image with the visual style of another. The technique was introduced by Leon Gatys, Alexander Ecker and Matthias Bethge in the paper A Neural Algorithm of Artistic Style, first posted in 2015 and later accepted at the CVPR conference in 2016; their method used a VGG-19 network that had been pretrained for object recognition. The algorithm generates a new image by minimizing two loss terms at once, one measuring how well the output preserves the subject matter of the content image and one measuring how well it matches the visual texture of the style image, backpropagating through the fixed network over many iterations to update the output image pixels directly. The technique has been widely adopted in creative applications, including mobile apps such as DeepArt and Prisma that transform photographs into images resembling the style of famous paintings. This description is adapted from Wikipedia contributors under CC BY-SA 4.0; changes were made. https://creativecommons.org/licenses/by-sa/4.0/