Facemorphing using stylegan2 and linear interpolation in w space

Notebooks

notebook.ipynb

Basic facemorph of two wild images using projection into W.

align_faces.ipynb

Face alignment batch script for aligning images like the FFHQ dataset.

batch_notebook.ipynb

Generate mulitple facemorph strips.

multi_morph_notebook.ipynb

Morph several images into one and display the process as a pyramid. This uses a brack style intepolation until one projection remains.

Util functions

Functions to call for generating facemorphs and aligning wild images.

UI

app.py

Simple ui for doing facemorph between two faces.

Get started

Clone the repo and inside it clone the stylegan2-ada-pytorch repo

create environment:

conda create -n "facemorphing" python=3.8.5

conda activate facemorphing

conda install click requests tqdm ninja imageio-ffmpeg==0.4.3

conda install imageio matplotlib

conda install numpy=1.24

Install pytorch using https://pytorch.org/get-started/locally/

Download 256 resolution pretrained model: https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada-pytorch/pretrained/transfer-learning-source-nets/ffhq-res256-mirror-paper256-noaug.pkl