arruw / fri-2021-masters-kindb-stylegan2-projector

Projecting KinDB to the StyleGAN2 latent space

Requirements

• Docker
• NVIDIA Container Runtime
• Conda (see ./conda-env.yml)
• NVIDIA GPU, NVIDIA drivers (see NVlabs/stylegan2-ada-pytorch)

Steps

• Clone this repository

$ git clone https://github.com/matjazmav/fri-2021-masters-kindb-stylegan2-projector.git
$ cd fri-2021-masters-kindb-stylegan2-projector/
$ git submodule update --init

• Build StyleGAN2 Docker image

$ tools/stylegan2-build.sh

• Create Conda env

$ conda env create -f conda-env.yml

• Set Google Drive credentials in file creds.json
• Download KinDB dataset from Google Drive

$ mkdir .cache
$ python gdrive.py download 1o_uH84Jk1GbEi0qHkIq2upNc8lZ6f6ex .cache/kindb.zip
$ unzip .cache/kindb.zip -d .cache/kindb

• Set .env file

# This configs can be used to split processing between multiple computers
REMINDER=0
MOD=1

• Run projector

$ nohup python projector.py &
$ echo $! > nohup.pid