Free daily packs with unique cards created by a DCGAN trained on 11,000+ real Pokemon cards. Trained on my laptop so images are not the crispiest :)
Hi 👋, my name is Oliver! I built this Pokemon card generator using a DCGAN trained on 11,000+ real cards. At this time of making this, I was very interested in Deep learning and had looked at the DCGAN Architecture paper. I also around the same time discovered the dataset of all pokemon card images. Combining the two, I decided I could make my own pokemon cards in this fusion of two interests!
This uses a Deep Convolutional Generative Adversarial Network (DCGAN). Two neural networks compete: one generates fake cards, the other tries to detect them. Reward functions are mapped based on how deceptful and how accurate a guess was. After thousands of rounds, the generator starts getting good enough to create realistic looking pokemon cards!
The model learned from 11,044 Pokemon TCG cards from the github Pokemon TCG Dataset. All cards were preprocessed to 600x825 pixels and filtered to only include traditional rectangular designs (no full-art cards). I removed all EX, MEGA and other cards.
