Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/jeremyfix/fakestylegan

Last synced: 5 days ago
JSON representation

Host: GitHub
URL: https://github.com/jeremyfix/fakestylegan
Owner: jeremyfix
Created: 2021-10-26T19:55:19.000Z (about 3 years ago)
Default Branch: main
Last Pushed: 2023-02-04T17:17:41.000Z (almost 2 years ago)
Last Synced: 2023-04-27T07:40:39.679Z (over 1 year ago)
Language: Python
Homepage:
Size: 1.85 MB
Stars: 2
Watchers: 2
Forks: 0
Open Issues: 0
Metadata Files:
- Readme: README.md

Awesome Lists containing this project

README

# Fakestylegan

Code to experiment with stylegan.

Expected target:

- [ ] stylegan running as a server on a GPU node,
- [x] waiting for input images and aligning it
- [ ] backproejcting onto the latent space
- [ ] playing around with the latent direction to generate new images

## Installation

python3 -m pip install git+https://github.com/jeremyfix/fakestylegan

You also need to have the `dnnlib` and `torch_utils` directories from the original [stylegan3](https://github.com/NVlabs/stylegan3) repository.

Then, if you are using a GPU, you need to set the `CUDA_HOME` variable appropriately :

```
export CUDA_HOME=/usr/local/cuda-11
```

and then have fun without our scripts, for example

python3 -m fakestylegan.generator

## Testing

### Alignment

For testing the alignment code, which is the one used by [NVlabs](https://github.com/NVlabs/ffhq-dataset/blob/master/download_ffhq.py) for realigning their input face image, you can :

python3 -m fakestylegan.align mysourceimage.jpg

### Interpolation in the latent space

The video [faces.avi](./examples/faces.avi) has been generated with `python3 -m fakestylegan.generator` and goes around in the
latent space to illustrate its topology and the incredible power of stylegan to generate faces.

[![See stylegan3 in action](https://img.youtube.com/vi/xNXCXO3LpEI/hqdefault.jpg)](https://youtu.be/xNXCXO3LpEI)

The interpolation scripts takes 30 seconds on a Geforce 3090, generating 200 images.

## References

Interesting papers are :

R. Abdal, Y. Qin, P. Wonka (2019). Image2StyleGAN : How to embed images into the styleGAN latent space. https://arxiv.org/pdf/1904.03189.pdf