Logo Generation with GAN
The goal of this project was to develop a Generative Adversarial Network (GAN) capable of generating unique logotypes for teams on a gaming platform. The trained GAN architecture enabled the creation of 1M unique logotypes for the players of the platform.
For this project, I was requested to deliver 1M logotypes for teams created on a gaming platform. I have researched and proposed a Generative Adversarial Network approach in order to create logotypes in an automatic fashion.
I leveraged a state-of-the-art approach incorporating transfer learning. I used the official TensorFlow implementation of StyleGAN2 with adaptive discriminator augmentation. The official pre-trained architecture was then tuned using the data exhibiting the style desired by the client for the project. The data was collected from various publicly available data sources. Such fine-tuned GAN architecture allowed me to automatically generate and deliver the requested logotypes.
Resources:
