VanGaugan

Multiple Convolutionnal Generative Adversarial networks implementations using Pytorch.

There are :

• Vanilla GAN - (dataset : MNIST)

• Conditionnal GAN - (dataset : MNIST, labels : number value)

• Deep Convolutionnal GAN - (datasets : CelebA & MNIST)

• Deep Convolutionnal Wassertein GAN - (CelebA, labels : gender)

The aim of the project is to create an Artificial Intelligence able to generate fake images, such as faces or landscapes, which appear real from a human point of view.

What is a GAN ?

GAN stand for Generative Adversarial Networks.

Here is a short definition that can be found on the internet.

Generative adversarial networks (GANs) are algorithmic architectures that use two neural networks, pitting one against the other (thus the “adversarial”) in order to generate new, synthetic instances of data that can pass for real data. They are used widely in image generation, video generation and voice generation.

(source)

GAN diagram representation

Installation

You can easily test the Deep Convolutional Generative Adversarial Network implementation.

$ git clone git@github.com:PoCInnovation/VanGaugan.git
$ pip3 install -r requirements.txt

Train a model

$ ./VanGaugan train -e [EPOCH_NUMBER] -g [GENERATOR_MODEL_SAVE_PATH] -d [DISCRIMINATOR_MODEL_SAVE_PATH]

GPU instance is required to train.

Load a model and display a grid of generated images

$ ./VanGaugan load [GENERATOR_MODEL_PATH]

Create a GIF to vizualise training progress

$ ./VanGaugan gif -o [GIF_OUTPUT_PATH] -md [MODELS_DIRECTORY_PATH]

Serve models through an HTTP API

$ ./VanGaugan serve

API endpoints

• GET /api/list-models : list availables models
• GET /api/<model_name>?image_number=<image_number>&label=<label> : generate images

Run front-end

$ cd front
$ npm install
$ npm run start

or create production build using npm run build and serve it whith what you want.

You can also use Docker : please take a look to Docker section.

Docker

docker-compose up

and wait a bit. You should be able to reach front-end here. It looks like this :

VanGaugan front-end screenshot

<u>N.B. :</u> Labels field has been added for a futur iteration

Docker architecture

Diagram representing VanGaugan docker architecture

Authors

• Clément Doucy
• Matthis Cusin