Abstract

Lecture posters are often used to promote a lecture. They can provide the information about a lecture’s theme, date, and etc. However, it’s difficult to generate lecture posters automatically, since we need to keep the design of them aesthetic and reasonable. In this paper, we study on how to generate a lecture poster from a background image and a text set. What we mainly need to consider is how to find an appropriate place on the background for a given text. In our method, we design two neural networks to determine the importance of a place on the background image and its harmony with text. Our generator is to arrange all the texts in the appropriate place based on some design guidelines with help of the two networks. To train our model, we create two datasets from some background images and well-designed posters collected online. The final results of the quality inspection experiments of the posters we generated show that our method is effective.

Author

       Yuncong Liu        Haoyu Ling         Luoxuan Li

Code

Our source code is available on Github.

Paper

Our paper is available here.

Video

Examples of results


poster generated by our method  poster generated by a simple generator