Generative-Adversarial-Network
-
Generative-Adversarial-Network
-
gan
-
GAN_debugging.ipynb
-
GAN_main.ipynb
-
README.md
-
download_celeba.sh
-
gan_sample.jpg
-
report.pdf
README.md
Generative Adversarial Network
I train a generative adversarial network (GAN) with spectral normalization of discriminator weights on the CelebA Dataset and generate facial images. The generator and discriminator network architectures I implement are roughly based on DCGAN [2]. I used GAN_debugging.ipynb notebook for debugging propuses. This notebook provides a small network you can use to train on MNIST. The small network trains very quickly so use it to verify that loss functions and training code are correct. All calculations were run on Google colab and Google Cloud.
Spectral normalization[1] helps improve the quality of generator images.
