Attribute-conditioned Layout GAN for Automatic Graphic Design

  • 2020-09-11 08:34:17
  • Jianan Li, Jimei Yang, Jianming Zhang, Chang Liu, Christina Wang, Tingfa Xu
  • 30

Abstract

Modeling layout is an important first step for graphic design. Recently,methods for generating graphic layouts have progressed, particularly withGenerative Adversarial Networks (GANs). However, the problem of specifying thelocations and sizes of design elements usually involves constraints withrespect to element attributes, such as area, aspect ratio and reading-order.Automating attribute conditional graphic layouts remains a complex and unsolvedproblem. In this paper, we introduce Attribute-conditioned Layout GAN toincorporate the attributes of design elements for graphic layout generation byforcing both the generator and the discriminator to meet attribute conditions.Due to the complexity of graphic designs, we further propose an element dropoutmethod to make the discriminator look at partial lists of elements and learntheir local patterns. In addition, we introduce various loss designs followingdifferent design principles for layout optimization. We demonstrate that theproposed method can synthesize graphic layouts conditioned on different elementattributes. It can also adjust well-designed layouts to new sizes whileretaining elements' original reading-orders. The effectiveness of our method isvalidated through a user study.

 

Quick Read (beta)

loading the full paper ...