Gradient Penalty DCGAN
Gradient Penalty DCGAN built on Keras.
gp-gan.py uses DCGAN architecture. gp-gan-res.py uses Residual Blocks with DCGAN. ggr-alpha.py uses AlphaGAN architecture with Residual Blocks.
Data collected from here: https://www.reddit.com/r/flowers
Related Papers: "Which Training Methods for GANs do actually Converge?" https://arxiv.org/abs/1801.04406
"Variational Approaches for Auto-Encoding Generative Adversarial Networks" https://arxiv.org/abs/1706.04987
To use, rename all images in your dataset according to this convention: "im (n).png". On windows you can do this by selecting all images, and renaming the first to "im". Then put all images into a folder under /data/. Change gp-gan-res.py to fit what you want (image size, directory, etc.) Run gp-gan-res.py.