Deep Convolutional Generative Adversarial Networks with Spectral Normalization using CelebA dataset.

• System with Linux or OSX
• NVIDIA GPU is strongly recommended for speed however CPU is supported but training will be slow.

1. sh setup.sh
2. python main.py --num_epochs 5 (Use -h for help)

You may run the program using DCGAN-with-SN.ipynb as well.

Number of Epochs	Fake Images	FID Score
20		116.44
100		34.74

• Pytorch DCGAN Tutorial: https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html
• https://github.com/christiancosgrove/pytorch-spectral-normalization-gan
• https://github.com/martinarjovsky/WassersteinGAN

• Vanilla GAN by Goodfellow et al.
• Deep Convolutional GAN by Radford et al.
• Spectral Normalization GAN by Miyato et al.
• Fréchet Inception Distance (FID) score Heusel et al.