Very simple program to create evolutionary art using CPPNs and a genetic algorithm.
Note: the program will run much faster with a cuda compatible GPU, but a CPU is still usable.
Clone or download the repository, then run main.py:
python3 main.py
A window will appear with a grid of generated images. To start creating art, select some images by clicking on them, then click on the next button (top right) to generate a new population of images based on your selected images from the previous population. Repeat this process until you like the images.
At any moment the animate button (top left) can be clicked to animate the grid of images currently displayed.
Note: If no images are selected and the next button is clicked, a new random population will be generated.
When you are happy with the results, you can select the images you want to save (same as before, by clicking on them), and click on the Save selected button (top middle). The saved images will go to a folder named "saved/" and will be in the ".png" format. The resolution of the saved images can be changed in the settings.py file (see Configuration).
The settings.py file contains all the configuration and can be changed to obtain different results. For example, the population_size parameter can be decreased to speed up the generation on CPU, or the saved_image_resolution parameter can be modified to change the output images resolution. Symetry can be added to the images by changing either horizontal_symetry or vertical_symetry, or both.
Here are the defaults:
### genetic algorithm settings
mutation_probability = 0.5
population_size = 25
### graphics settings
window_size = (620, 720)
image_size = 100
image_resolution = 100
channels = 1 # 1 for black/white or 3 for colors
padding = 20
grid_size = 5
grid_offset = (0, 100)
max_coord_scale = 3
saved_image_resolution = 1000
### animation settings
time_scale = 10
time_step = 0.2
current_time = 1.0
### model settings
nb_neuron_per_layer = 15
init_nb_layers = 2
nb_input_params = 6
horizontal_symetry = False
vertical_symetry = False
weights_mean = 0.0
max_weights_std = 20
latent_std = 10
### pytorch settings
import torch
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
### debug settings
show_generation_time = False