bshantam97/GAN-Architectures

GAN-Architectures

Author

Shantam Bajpai

Architectures Implemented

1. DCGAN (Deep Convolutional Generative adversarial network)

2. WGAN (Wasserstein Generative Adversarial Network)

3. WGAN With Gradient Penalty

4. CGAN (Conditional Generative adversarial network)

5. EBGAN (Energy Based Generative Adversarial Network)

To Do

Calculate the Inception Scores and Compute the Fretchet Inception Distance.

Dataset Used

The dataset used to train the Generative adversarial networks was the celeba dataset which is a large scale face attributes dataset with more than 200K Celebrity faces and the MNIST Dataset (For conditional Wasserstein GAN-GP).

Tensorboard Visualizations

Fake Images Generated using DCGAN

Loss curves for DCGAN

Fake Images Generated using WGAN for 5 epochs

Loss Curves for WGAN (5 Epochs)

Fake Images generated using WGAN-GP for 5 epochs

Loss Curves for WGAN-GP

Fake Images generated using Conditional WGAN-GP after training for 20 Epochs

Loss Curves for Conditional WGAN-GP

Research Paper References

DCGAN: https://arxiv.org/pdf/1511.06434.pdf

WGAN: https://arxiv.org/pdf/1701.07875.pdf

WGAN-GP: https://arxiv.org/pdf/1704.00028v3.pdf

CGAN: https://arxiv.org/pdf/1411.1784.pdf

EBGAN: https://arxiv.org/pdf/1609.03126.pdf