/SalS-GAN

Primary LanguagePythonMIT LicenseMIT

SalS-GAN

The code for our paper SalS-GAN: Spatially-Adaptive Latent Space in StyleGAN for Real Image Embedding.
Our code is heavily borrowed from Idinvert. Feel free to raise any issues, we will reply to them as soon as possible.

Inversion Example

Step-1:Get Pretrained Models

Download pre-trained models(StyleGAN models and VGG model) and put them in "./models/pretrain". You can find some pretrained stylegan models in Idinvert.

Step-2:Invert Image

python invert.py $MODEL_NAME $IMAGE_LIST -o $OUTPUT_DIR --num_iterations ITERATIONS  
example: python invert.py styleganinv_tower256  tower_val_256x256/test.list -o ./outputs/tower  --num_iterations 3000

Citation

@inproceedings{zhang2021salsgan,  
  title={SalS-GAN: Spatially-Adaptive Latent Space in StyleGAN for Real Image Embedding},  
  author={Lingyun Zhang, Xiuxiu Bai, and Yao Gao},  
  booktitle={Proceedings of the 29th ACM International Conference on Multimedia (MM ’21)},  
  year={2021}  
}