GAN

Using a Generative Adversarial Network to Generate Images

Neural Network Model

A GAN consists of two Neural Networks a Generative Network and a Discriminative Networks.

• Generative Network: It is a network which takes in a random noise vector and generates images that are "similar" to the input images.

• Adversarial Network: It is a network which takes in "Fake" and Real images, and acts as a binary classifier, classifying whether the input is fake or real.

Together the networks will try to decrease their respective losses, hence the name "Adverserial Networks"

Requirements:

• Python 3.6.2 (https://www.python.org/downloads/release/python-362/)
• Numpy (https://pypi.org/project/numpy/)
• Tensorflow (https://pypi.org/project/tensorflow/)
• Keras (https://pypi.org/project/Keras/)
• OpenCV (https://pypi.org/project/opencv-python/)

Usage:

• Download the required images, and put it in the data folder:

Note: To get the best result ensure that the images don't vary dratically

• To preprocess the images run resize.py, followed by RGBA2RGB.py with the following argumanets

usage: resize.py [-h] --input INPUT --output OUTPUT

Resize Input Images

optional arguments:
  -h, --help       show this help message and exit
  --input INPUT    Directory containing images to resize. eg: ./data
  --output OUTPUT  Directory to save resized images. eg: ./resized

usage: RGBA2RGB.py [-h] --input INPUT --output OUTPUT

Convert RGBA to RGB

optional arguments:
  -h, --help       show this help message and exit
  --input INPUT    Directory containing images to resize. eg: ./resized
  --output OUTPUT  Directory to save resized images. eg: ./RGB_data

• Now you can run the GAN on the final processed images, using GAN.py

usage: GAN.py [-h] --mode MODE [--name NAME] [--input INPUT] [--output OUTPUT]
              [--epoch EPOCH] [--batch BATCH]

Train or Test the Generative Adverserail Network

optional arguments:
  -h, --help       show this help message and exit
  --mode MODE      Whether to Test or Train
  --name NAME      Directory of the Generated Images eg: NewPaints
  --input INPUT    Directory of input Images eg: RGB_data
  --output OUTPUT  Output Image Name
  --epoch EPOCH    Number of Epochs to Run
  --batch BATCH    Batch Size

• To crop out individual images from the generated batch you can use crop.py

usage: crop.py [-h] --input INPUT --output OUTPUT

Crop Individual Images

optional arguments:
  -h, --help       show this help message and exit
  --input INPUT    Input Image
  --output OUTPUT  Output Directory

Generated Examples

The GAN was trained for 29 hrs on a dataset consisting of various paintings of landscapes, and it came up with these

These aren't perfect because perfecting the generator requires a lot of computational power.

G00D LUCK

For doubts email me at: [email protected]