Alexander Sage

Semester Work
Supervisors: Eirikur Agustsson, Dr. Radu Timofte

GAN Training on Multi-Modal Data with Application for Logo Synthesis

Designing a logo for a new brand usually is a lengthy and tedious back-and-forth process between a designer and a client. The goal of this project is to explore to what extent, artificial intelligence can solve the creative task of the designer. Towards this goal, we build a dataset of 500k+ logos crawled from the world wide web (https://data.vision.ee.ethz.ch/cvl/lld). Training Generative Adversarial Networks (GANs) on this multi-modal data is not straightforward, resulting in mode collapse for the most recent state of the art methods. To address this, we introduce techniques to stabilize the GAN training and obtain promising preliminary results.