ICIP2023

0 Enviroments

pip install -r requirement_new.txt

1 To use the demo

refer to TLGAN https://github.com/SummitKwan/transparent_latent_gan

Decompress the downloaded files and put it in project directory as the following format, which can be decompressed in transparent_latent_gan_ICIP.7z.

```text
root(d):
  asset_model(d):
    karras2018iclr-celebahq-1024x1024.pkl   # pretrained GAN from Nvidia
    network-final_yitu.pkl
    network-snapshot-007286.pkl
    network-snapshot-008486.pkl
    cnn_face_attr_celeba(d):
      model_20180927_032934.h5              # trained feature extractor network
    dnnlib(d)
  asset_results(d):
    pg_gan_celeba_feature_direction_40(d):
      feature_direction_20181002_044444.pkl # feature axes
    pg_gan_celeba_feature_direction_celebA
    pg_gan_celeba_feature_direction_celebA_givens
    pg_gan_celeba_feature_direction_celebA_house
    pg_gan_celeba_feature_direction_celebA_smitt
    pg_gan_celeba_feature_direction_yitu_givens
    pg_gan_celeba_feature_direction_yitu_house
    pg_gan_celeba_feature_direction_yitu_smitt
    stylegan_ffhq_explore
    stylegan_ffhq_feature_direction_40
    stylegan_ffhq_feature_direction_retrained
    
    
```

2 To use the demo

run the interactive demo from the Jupyter Notebook at ./src/notebooks/tl_gan_ipywidgets_gui_celebA.ipynb

feature direction can be changed by change "path_feature_direction"