Skip to content

PyTorch implementation of 'Conditional GAN' (Mirza et al., 2014) & 'WGAN-GP' (Gulrajani et. al., 2017) and training it on MNIST

Notifications You must be signed in to change notification settings

KimRass/Conditional-WGAN-GP

Repository files navigation

1. Pre-trained Model

  • cwgan_gp_mnist.pth
    seed=888
    n_epochs=50
    batch_size=64
    lr=0.0002
    d_hidden_dim=32
    g_latent_dim=100
    g_hidden_dom=32
    gp_weight=10
    n_d_updates=3

2. Samples

3. Implementation Details

1) Architecture

  • [1]에서 Architecture를 가져와서 몇 가지를 변경했습니다.
    • Discriminator:
      • 첫 번째 Convolutional layer 다음에 Batch normalization layer를 추가했습니다.
    • Generator:
      • 마지막 Transposed convolutional layer에서 bias=True로 변경했습니다.
      • ReLU activation을 Leaky ReLU activation으로 변경했습니다.
  • 이렇게 변경함으로써 샘플의 퀄리티가 상승했습니다.

4. References

About

PyTorch implementation of 'Conditional GAN' (Mirza et al., 2014) & 'WGAN-GP' (Gulrajani et. al., 2017) and training it on MNIST

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published