Generative Adversarial Networks - GANs

This repository presents the basic notions that involve the concept of Generative Adversarial Networks.

"...the most interesting idea in the last 10 years in ML". Yann LeCun

Definition

Generative Adversarial Networks or GANs is a framework proposed by Ian Goodfellow, Yoshua Bengio and others in 2014.

GANs are composed of two models, represented by artificial neural network:

• The first model is called a Generator and it aims to generate new data similar to the expected one.
• The second model is named the Discriminator and it aims to recognize if an input data is ‘real’ — belongs to the original dataset — or if it is ‘fake’ — generated by a forger.

Read more in this post GANs — Generative Adversarial Networks 101.

Configure environment

• Create the conda environment

(base)$: conda env create -f environment.yml

• Activate the environment

(base)$: conda activate gans_101

• Run!

(gans_101)$: python -m jupyter notebook

⚠️ Note

🎉 Run with conda environment:

• GAN - MNIST

🚧 Under construction:

• DCGAN - MNIST
• CGAN - MNIST
• CCGAN - MNIST
• WGAN - MNIST
• LSGAN - MNIST
• DCGAN - CIFAR10
• CGAN - CIFAR10

Models

Definition and training some models with MNIST and CIFAR-10 datasets.

MNIST dataset

• GAN - MNIST Notebook - Post Medium
• DCGAN - MNIST Notebook - Post Medium
• CGAN - MNIST Notebook - Post Medium
• CCGAN - MNIST Notebook - Post Medium
• WGAN - MNIST Notebook - Post Medium
• LSGAN - MNIST Notebook - Post Medium

CIFAR-10 dataset

• DCGAN - CIFAR10 Notebook - Post Medium
• CGAN - CIFAR10 Notebook - Post Medium

Results

Training models with Keras - TensorFlow.

MNIST dataset

Generative Adversarial Networks - GANs

A GANs implementation using fully connected layers. Notebook

Epoch 00	Epoch 100	Loss

Deep Convolutional Generative Adversarial Networks - DCGANs

A DCGANs implementation using the transposed convolution technique. Notebook

Epoch 00	Epoch 100	Loss

Conditional Generative Adversarial Nets - CGANs

A CGANs implementation using fully connected layers and embedding layers. Notebook

Epoch 00	Epoch 100	Loss

Context-Conditional Generative Adversarial Networks - CCGANs

A CCGANs implementation using U-Net and convolutional neural network. Notebook

Epoch 00	Epoch 100	Loss

Wasserstein Generative Adversarial Networks - WGANs

A WGANs implementation using convolutional neural network. Notebook

Epoch 00	Epoch 100	Loss

Least Squares General Adversarial Networks - LSGANs

A LSGANs implementation using using fully connected layers. Notebook

Epoch 00	Epoch 100	Loss

CIFAR-10 dataset

Deep Convolutional Generative Adversarial Networks - DCGANs

A DCGANs implementation using the transposed convolution technique. Notebook

Epoch 00	Epoch 100	Loss

Conditional Generative Adversarial Networks - CGANs

A CGANs implementation using the transposed convolution and convolution neural network, and concatenate layers. Notebook

Epoch 00	Epoch 100	Loss

References

• Complete Post Medium

  • GANs — Generative Adversarial Networks 101

• Related papers:

  • Generative Adversarial Networks
  • Unsupervised Representation Learning With Deep Convolutional
  • Conditional Generative Adversarial Nets
  • Semi-Supervised Learning with Context-Conditional Generative Adversarial Networks
  • Wasserstein GAN
  • Least Squares General Adversarial Networks

• Datasets:

  • THE MNIST DATABASE of handwritten digits
  • The CIFAR-10 dataset

• Other repositories:

  • Keras-GAN
  • Adversarial Nets Papers
  • How to Train a GAN? Tips and tricks to make GANs work
  • The GAN Zoo
  • GAN Lab: An Interactive, Visual Experimentation Tool for Generative Adversarial Networks
  • gans-awesome-applications
  • tensorflow-generative-model-collections

made with 💙 by mafda