This repo provides codes for "Structured Generative Adversarial Networks" and is based on TripleGAN.

For example, you can run python -u sgan_cifar10.py -ssl_seed 1 to reproduce the semi-supervised classification results on CIFAR-10 dataset.

You can run python -u generate.py -oldmodel ... to generate samples, infer latent codes and transfer image styles based on a trained model.