Style is a Distribution of Features

  • 2020-07-25 21:17:51
  • Eddie Huang, Sahil Gupta
  • 47

Abstract

Neural style transfer (NST) is a powerful image generation technique thatuses a convolutional neural network (CNN) to merge the content of one imagewith the style of another. Contemporary methods of NST use first or secondorder statistics of the CNN's features to achieve transfers with relativelylittle computational cost. However, these methods cannot fully extract thestyle from the CNN's features. We present a new algorithm for style transferthat fully extracts the style from the features by redefining the style loss asthe Wasserstein distance between the distribution of features. Thus, we set anew standard in style transfer quality. In addition, we state two importantinterpretations of NST. The first is a re-emphasis from Li et al., which statesthat style is simply the distribution of features. The second states that NSTis a type of generative adversarial network (GAN) problem.

 

Quick Read (beta)

loading the full paper ...