Code to fully reproduce the results for the blog post These Bored Apes Do Not Exist.
View results at thisboredapedoesnotexist.nathancooperjones.com.
On a machine with Python installed, first install requirements with:
pip install -r requirements.in
From here, you can take one of two paths:
-
Training a model from scratch. To do this, you'll need some data. With a free GPU available and plenty of time, head on over to
00_get_data
to get the dataset needed for the GAN. From there, you'll head to01_lightweight_gan
,02_super_resolution
, and03_resizing_and_png_conversion
, in that order. -
Running pre-trained models in inference-mode. If you do not want to train any models, but instead just run all models in inference mode, then head over to
01_lightweight_gan
to find the URL to download the model weights. From there, you'll head to02_super_resolution
and03_resizing_and_png_conversion
, in that order.
This project would not be possible without:
- Phil Wang's (AKA lucidrains) incredible work on Lightweight GAN!
- yu45020's understandable port of Waifu2x!
- Inspiration, motivation, and encouragement from Michael Sugimura (AKA sugi-chan on both GitHub and Medium).
- My partner Jenna Blazi, who let me ramble about NFTs and this project for two weeks straight and (almost) never complained about it.