Awesome

gans: Generative Adversarial Networks

Multiple Generative Adversarial Networks (GANs) implemented in PyTorch and Tensorflow.

Check out this blog post for an introduction to Generative Networks.

<img src=".images/dcgan_mnist.gif" width="275"> <img src=".images/dcgan_cifar.gif" width="275">

Vanilla GANs

Vanilla GANs found in this project were developed based on the original paper Generative Adversarial Networks by Goodfellow et al.

These are trained on the MNIST dataset, and learn to create hand-written digit images using a 1-Dimensional vector representation for 2D input images.

• PyTorch Notebook
• TensorFlow Notebook

<img src=".images/vanilla_mnist_pt_raw.png" width="300"> <img src=".images/vanilla_mnist_pt.png" width="300">

MNIST-like generated images before & after training.

DCGANs

Deep Convolutional Generative Adversarial Networks (DCGANs) in this repository were developed based on the original paper Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks by Radford et al.

These are trained on the CIFAR10 and the MNIST datasets. They use 3 dimensional representations for images (length x height x colors) directly for training.

• TensorFlow CIFAR10 Notebook
• PyTorch CIFAR10 Notebook
• PyTorch MNIST Notebook

<img src=".images/dcgan_cifar_pt_raw.png" width="300"> <img src=".images/dcgan_cifar_pt.png" width="300">

CIFAR-like generated images before & after training.