Aesthetics++: Refining Graphic Designs by Exploring Design Principles and Human Preference
Wenyuan Kong,Zhaoyun Jiang,Shizhao Sun,Zhuoning Guo,Weiwei Cui,Ting Liu,Jianguang Lou,Dongmei Zhang
DOI: https://doi.org/10.1109/tvcg.2022.3151617
IF: 5.2
2023-01-01
IEEE Transactions on Visualization and Computer Graphics
Abstract:During the creation of graphic designs, individuals inevitably spend a lot of time and effort on adjusting visual attributes (e.g., positions, colors, and fonts) of elements to make them more aesthetically pleasing. It is a trial-and-error process, requires repetitive edits, and relies on good design knowledge. In this work, we seek to alleviate such difficulty by automatically suggesting aesthetic improvements, i.e., taking an existing design as the input and generating a refined version with improved aesthetic quality as the output. This goal presents two challenges: proposing a refined design based on the user-given one, and assessing whether the new design is better aesthetically. To cope with these challenges, we propose a design principle-guided candidate generation stage and a data-driven candidate evaluation stage. In the candidate generation stage, we generate candidate designs by leveraging design principles as the guidance to make changes around the existing design. In the candidate evaluation stage, we learn a ranking model upon a dataset that can reflect humans’ aesthetic preference, and use it to choose the most aesthetically pleasing one from the generated candidates. We implement a prototype system on presentation slides and demonstrate the effectiveness of our approach through quantitative analysis, sample results, and user studies.