Logo GANerator

Wedding Dress
Wedding Dress Polished
Little Girl in Red
Wandering Hero
Dropkickin' Sloth
Fire Mage
Creepy Owl
Example of bad GAN output

Inspiration

One of our team members read about Generative Adversarial Networks (GAN) in his research seminar, so we looked for a new domain to apply GANs to. We decided to pursue creating logos because the training files were easy to obtain, and the website would be useful, as thousands of start-ups are created everyday that want to find a logo design.

What it does

From a bunch of existing company logos, a model is built that can assemble pieces of the existing logos together. The output of the model is loaded onto a website when a user requests to see more logo designs.

How I built it

The GAN network was taken from https://github.com/carpedm20/DCGAN-tensorflow. This originally ran on face data, and we modified the code and data parameters so the algorithm worked for logos.

Challenges I ran into

Fixing some bugs of the pre-existing GAN software.

Accomplishments that I'm proud of

Getting the GAN to run. Polishing GAN outputs with Paint.net for clearer visualization and as a demonstration of moving from logo prototype to possible actual logo.