README.md

Pytorch Implementation of SN-GAN with CIFAR10

Requirements

nsml (not neccessary) visdom

Paper

Spectral Normalization for Generative Adversarial Networks

Run Example

If you have nsml:

nsml run -d cifar10_python -a "--sn"                   # Spectralnorm
nsml run -d cifar10_python                             # No SN
nsml run -d cifar10_python -a "--sn --inception_score" # Calculate Inception score

If you don't:

python main.py --sn                    # Spectralnorm
python main.py                         # No SN
python main.py --sn --inception_score" # Calculate Inception score

Architecture

GAN Architecture is adopted from the papers' Appendix B.4 for CIFAR10

Results

Generated images

Loss