LukasDrews97 / gan_evaluation

Synthetic Image Generation using Generative Adversarial Networks (GANs)

Synthetical Image Generation using Generative Adversarial Networks (GANs)

In this project, I generated synthetical images using Non Saturating GANs (NSGANs), Wasserstein-GANs (WGANs) and Deep-Convolutional GANs (DCGANs). Each model was evaluated using the following metrics:

• Inception Score
• Fréchet Inception Distance
• Kernel Inception Distance
• Unbiased Inception Score & Unbiased Fréchet Inception Distance (https://arxiv.org/pdf/1911.07023v3.pdf)

Using the following datasets:

• MNIST
• Fashion-MNISTs
• Cifar-10
• CelebA

Project Structure

File/Folder	Description
models	Folder containing the GAN models
train_dcgan.py	Entry point for training a DCGAN
train_nsgan.py	Entry point for training a NSGAN
train_wgan.py	Entry point for training a WGAN
config	Folder containing config files for determined.ai
metrics	Folder containing implemented metrics
datasets.py	Contains all datasets for training

Insights