kingofprank / SGF

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

SGF

Unofficial PyTorch implementation of Surrogate Gradient Field for Latent Space Manipulation

Required

  • PyTorch
  • dlib
  • OpenCV

Pretrained Checkpoints

Inference

Run python translate_keypoints.py. This script aligns the face keypoints of a randomly generated image with another randomly generated image.

Train on your own dataset

  1. Create a Classifier Network on models/classifier.py.
  2. Run make_dataset.py to create a dataset by generating images and classifiering their properties.
    python make_dataset.py \
         --size [images size of stylegan2] \
         --g_ckpt [the pretrained checkpoint of stylegan2] \
         --data_dir [where you want to save training data] \
         --n_sample 200000 \
         --batch 8 \
         --truncation 0.8
    
  3. Run train.py.
    python train.py \
         --total_iter 500000 \
         --data_dir data \
         --batch 8 \
         --n_layer 15
    

License

StyleGAN2 codes and model are from rosinality/stylegan2-pytorch.

Face-Parsing codes and checkpoint are from Zllrunning/face-parsing.PyTorch.

senet.py is from cydonia999/VGGFace2-pytorch, and the face attributes classifier model was transfer trained from pretrained weights provided by the above repository.

About


Languages

Language:Python 83.4%Language:Cuda 11.6%Language:C++ 4.6%Language:C 0.4%