An unofficial PyTorch implementation of the paper Deep Interactive Evolution.
Image stolen from the paper.
This paper combines generative adversarial networks with interactive evolutionary computation. Specifically, instead of randomly sampling from gans, a user can guide the generation by selecting images with desired traits using an interactive gui.
The author of this repo does not possess the hardware, the time, the patience or the skills necessary to train gans. Threfore, the pretrained models from Facebook's GAN zoo are employed.
Whereas the authors of the paper developed a web interface to display images, the author of this repo possesses zero web development skills and therefore makes due with a makeshift tkinter gui.
$ git clone https://github.com/davidsvy/interactive-evolution
$ cd interactive-evolution
$ pip install -r requirements.txt
$ python run.py [-c configs/config.yaml]
@misc{bontrager2018deep,
title={Deep Interactive Evolution},
author={Philip Bontrager and Wending Lin and Julian Togelius and Sebastian Risi},
year={2018},
eprint={1801.08230},
archivePrefix={arXiv},
primaryClass={cs.NE}
}