Title:Retrieve-Then-Adapt: Example-based Automatic Generation for Proportion-related Infographics

Abstract:Infographic is a data visualization technique which combines graphic and textual descriptions in an aesthetic and effective manner. Creating infographics is a difficult and time-consuming process which often requires significant attempts and adjustments even for experienced designers, not to mention novice users with limited design expertise. Recently, a few approaches have been proposed to automate the creation process by applying predefined blueprints to user information. However, predefined blueprints are often hard to create, hence limited in volume and diversity. In contrast, good infogrpahics have been created by professionals and accumulated on the Internet rapidly. These online examples often represent a wide variety of design styles, and serve as exemplars or inspiration to people who like to create their own infographics. Based on these observations, we propose to generate infographics by automatically imitating examples. We present a two-stage approach, namely retrieve-then-adapt. In the retrieval stage, we index online examples by their visual elements. For a given user information, we transform it to a concrete query by sampling from a learned distribution about visual elements, and then find appropriate examples in our example library based on the similarity between example indexes and the query. For a retrieved example, we generate an initial drafts by replacing its content with user information. However, in many cases, user information cannot be perfectly fitted to retrieved examples. Therefore, we further introduce an adaption stage. Specifically, we propose a MCMC-like approach and leverage recursive neural networks to help adjust the initial draft and improve its visual appearance iteratively, until a satisfactory result is obtained. We implement our approach on proportion-related infographics, and demonstrate its effectiveness by sample results and expert reviews.