neural-net-random-art

Create a grayscale or colour image with predefined size img_height and img_width using fully connected neural networks.
The generation of images only requires python numpy, torch and matplotlib.
Medium article can be found here.

Usage

You can either have a look at the jupyter notebook nb_random_art.ipynb if you want to understand the algorithm and check out several settings of the method.
For fast image generation is is recommended to use the python main programm file random_art.py

Using Google Colab

I've created a minimal working example on Google Colab here here in case you want to experiment with this library a little bit, prior to installing it on your local machine.
To generate an animation by applying small gradual changes in the neural net when generating consecutive new images, check out this notebook.

Dependencies: Python 3

numpy
matplotlib
seaborn
torch

Installation

git clone https://github.com/tuanle618/neural-net-random-art.git
cd neural-net-random-art
pip install -e .

Execution

For the random_art.py program argparse is used to define several input parameters:

parser = argparse.ArgumentParser(description="Generate random art with a deep neural network")

parser.add_argument("-img_height", metavar="", type=int, default=512,
                    help="Image height of created random art. Default is 512")

parser.add_argument("-img_width", metavar="", type=int, default=512,
                    help="Image width of created random art. Default is 512")

parser.add_argument("-colormode", metavar="", type=str, default="rgb",
                    help="How image color should be generated. Options are ['bw', 'rgb', 'cmyk', 'hsv', 'hsl']."
                         " By default this value is 'rgb'")
    
parser.add_argument("-alpha", metavar="", type=str, default="False",
                    help="Whether or not to add a alpha channel for the image. Default is False")

parser.add_argument("-n_images", metavar="", type=int, default=1,
                    help="Number of images to generate. Default is 1")
    
parser.add_argument("-n_depth", metavar="", type=int, default=5,
                    help="Number of layers for the neural network. Default is 5")

parser.add_argument("-n_size", metavar="", type=int, default=10,
                    help="Number of neurons in each hidden layer. Default is 10")

parser.add_argument("-activation", metavar="", type=str, default="tanh",
                    help="Activation function to apply on the hidden layers. Default is 'tanh'")
    
parser.add_argument("-z1", metavar="", type=float, default=-0.618,
                    help="Input variable 1 to insert determinism into the random art."
                         " The value should be between -1 and 1. Default is -0.618")
    
parser.add_argument("-z2", metavar="", type=float, default=+0.618,
                    help="Input variable 2 to insert determinism into the random art."
                         " The value should be between -1 and 1. Default is +0.618")

parser.add_argument("-trig", metavar="", type=str, default="True",
                    help="If the z1 and z2 values should be transformed with cosine and sine respectively. "
                         "Defaults to True.")

parser.add_argument("-noise", metavar="", type=str, default="False",
                    help="If gaussian noise should be added for the generated image. Defaults to False")

parser.add_argument("-noise_std", metavar="", type=float, default=0.01,
                    help="Gaussian noise standard deviation if it should be added to the generated image. "
                         " Defaults to 0.01.")

parser.add_argument("-sym", metavar="", type=str, default="False",
                    help="Use symmetry network. Default is False")

parser.add_argument("-gpu", metavar="", type=str, default="False",
                    help="Use GPU to generate (vectorized) image. Defaults to False")

parser.add_argument("-format", metavar="", type=str, default="png",
                    help="File format to save the images. Defaults to 'png'."
                         "Choices are 'pnd', 'jpg', 'pdf' and 'svg'.",
                    choices=['png', 'jpg', 'svg', 'pdf']
                    )

So in order to create 1 RGB image of size 400x500, no alpha channel, a dense net with 15 layers, each laying having 15 neurons, type in following command in the shell:

python random_art.py -img_height 400 -img_width 500 -colormode rgb -alpha False -n_images 1 -n_depth 15 -n_size 15

Examples

Following commands were used [default params were used] to get the images stored in the result subdirectory:

python random_art.py -img_height 512 -image_width 512 -colormode bw -alpha False -n_images 5