pytorch-gan-zoo 0.0.6

Implementation of GAN models in PyTorch.

GAN-zoo

Train and evaluate basic FC-GAN:

# train on fashion-mnist
python -m ganzoo.examples.basic_gan.train_fc \
    --z_dim=100 --num_hidden_units=256 \
    --network_type=fc-large \
    --dataset_name=fashion-mnist

# evaluate
python -m ganzoo.examples.basic_gan.eval_fc \
    lightning_logs/version_1/checkpoints/epoch\=99-step\=337600.ckpt \
    --output_dir lightning_logs/version_1/outputs

A collection of GAN models implemented in PyTorch:

• Basic GAN on MNIST code
• Conditional GAN on MNIST code
• Pix2Pix on FACADES dataset code

References - part 1

• GAN arxiv
• Wasserstein GAN arxiv
• DCGAN arxiv
• Conditional GANs arxiv
• LSGAN arxiv
• SAGAN arxiv
• Pix2Pix arxiv
• Pix2Pix-HD arxiv PyTorch imp.
• CycleGAN arxiv code
• StarGAN arxiv PyTorch imp.
• Geometric GAN arxiv
• InfoGAN arxiv
• BigGAN arxiv
• ConSinGAN arxiv code
• OT-GAN arxiv

References - part 2

• Improved Techniques for Training GANs arxiv
• On Convergence and Stability of GANs arxiv
• Which Training Methods for GANs do actually Converge? arxiv
• Is generator conditioning causally related to GAN performance? arxiv
• The unusual effectiveness of averaging in gan training arxiv
• cGANs with projection discriminator arxiv