jimothyjohn / stylegan-streamlit

Official PyTorch implementation of StyleGAN3

StyleGAN-Scripts

Description

This is a Streamlit app for StyleGAN3. You'll want to go there for information since this is just a demo utility.

Prereqs

• Debian-based Linux
• NVIDIA GPU >=8GB
• CUDA >=11
• cuDNN >=8.2

Setup

Clone the repo and move on in:

sudo apt update && sudo apt -y install git
git clone https://github.com/JimothyJohn/stylegan-scripts # clone repo
cd stylegan-scripts # move to repo directory

Anaconda

Install Anaconda environment and initialize with:

conda env create -f environment.yml
conda activate stylegan

Docker

Build Docker image and run with:

docker/docker_build.sh
utils/Streamlit.sh

Model Zoo

Subject	Base	Resolution (px x px)	kimgs	FID	Link
Whataburger	StyleGAN3-R	256	1k	???	https://drive.google.com/file/d/1y6rNQr9lnYNtshQpeL0KvjkDStYwssIc/view?usp=sharing
Monet	StyleGAN3-R	256	1k	???	https://drive.google.com/file/d/1IKxpWjcuz0BWSaswyeXdUcC3ZCQxzCWM/view?usp=sharing
Galaxy	StyleGAN3-R	512	1k	70	https://drive.google.com/file/d/15QsTVf5CQhQqjWqq7_EWjFVP1Lgvi3FN/view?usp=sharing
WikiArt	StyleGAN3-R	1024	???	8.1	https://drive.google.com/file/d/18MOpwTMJsl_Z17q-wQVnaRLCUFZYSNkj/view?usp=sharing

To-do

• Unify synthesis into single page (mappings)
• Add progress bars to UI
• Add video preview for mixing
• Accelerate projection algorithm using either previous vectors OR extract generator encoder
• Make device agnostic (.cuda() -> .to(device))
• Automate dataset creation wiht fiftyone