Yu-Gi-Oh is a popular trading card game with a large variety of artworks. This makes for an interesting dataset for a data generation problem. In our attached paper, we propose a GAN architecture for generating images for Yu-Gi-Oh! cards.
In the attached paper, we describe our methodology and findings. We propose a novel GAN architecture to generate new images of Yu-Gi-Oh! cards. We also discuss the model's sucseptibility to mode collapse, why this is a challenge for this particular dataset, and methods for overcoming the mode collapse issue.
Please read the attached file YuGANOh paper.pdf for more info.
I have included the training jupyter notebook, but I cannot host the training data on GitHub. I suggest downloading the software YGOPro and ripping the loaded images from there.