GANForge

Python library for a wide variety of GANs (Generative Adversarial Networks) based on TensorFlow and Keras.

Table of Contents

1. Installation
2. Quick Start
3. Examples
4. Supported GANs
5. Custom Callbacks

Installation

To download the GANForge model from pypi please use the following pip command in your command prompt/terminal

pip install git+https://github.com/quadeer15sh/GANForge.git

Quick Start

You can get started with building GANs in just a few lines of code.

Example:

import tensorflow as tf
from GANForge.dcgan import DCGAN

# train_ds: your image dataset

model = DCGAN(input_shape=(64, 64, 3), latent_dim=128)
model.compile(d_optimizer=tf.keras.optimizers.Adam(learning_rate=0.0002),
              g_optimizer=tf.keras.optimizers.Adam(learning_rate=0.0002),
              loss_fn=tf.keras.losses.BinaryCrossentropy())

model.fit(train_ds, epochs=25)

Examples

Please feel free to explore through the notebook files on each of the GAN models available in GANForge

• Jupyter Notebooks
• Kaggle Implementations

Supported GANs

Note : This list is updated frequently, please come back to check if the GAN architecture you desire to use is available or not

Sr.	GAN Architecture	Status
1	DC GAN	Available
2	Conditional GAN	Available
3	Info GAN	In Progress
4	SR GAN	Available
5	ESR GAN	In Progress
6	Pix2Pix GAN	In Progress
7	Cycle GAN	In Progress
8	Attention GAN	In Progress

Custom Callbacks

Custom callbacks available for usage during your training

Sr.	Callback	GAN Applicable
1	DCGANVisualization	DC GAN
2	ConditionalGANVisualization	Conditional DCGAN

Example:

import tensorflow as tf
from GANForge.dcgan import DCGAN
from GANForge.callbacks import DCGANVisualization

# train_ds: your image dataset

model = DCGAN(input_shape=(64, 64, 3), latent_dim=128)
model.compile(d_optimizer=tf.keras.optimizers.Adam(learning_rate=0.0002),
              g_optimizer=tf.keras.optimizers.Adam(learning_rate=0.0002),
              loss_fn=tf.keras.losses.BinaryCrossentropy())
visualizer = DCGANVisualization(n_epochs=5)

model.fit(train_ds, epochs=25, callbacks=[visualizer])