Pytorch implementation of several GANs with conditional signals (supervised or unsupervised). All experiments are conducted on Fashion-MNIST, and the network structures are adapted from Improved GAN.

• Supervised
  • CGAN
  • CGAN with Projection Discriminator
  • ACGAN
  • Others
• Unsupervised
  • InfoGAN
  • Others

CGAN	Projection CGAN	ACGAN
InfoGAN1	InfoGAN2	InfoGAN3

• Prerequisites

  • PyTorch 1.0.0
  • Python 3.6

• Examples of training

  • training

    CUDA_VISIBLE_DEVICES=0 python train_CGAN.py

  • tensorboard for loss visualization

    CUDA_VISIBLE_DEVICES='' tensorboard --logdir ./output/CGAN_default/summaries --port 6006

• Others

  • If you want to use other datasets, just replace FashionMNIST by MNIST or CIFAR10 in the codes.
  • There are arguments for configurations of GAN loss, gradient penalty, and etc, just try them.