FLamefiREz/modulated_stargan

StarGAN V2 with Modulated Convolutions

StarGAN V2 with Modulated Convolutions

Experiments on replacing Adaptive Instance Normalization(AdaIN) layers in StarGAN V2 model.

The presented results was obtained in a 24 hours on a single Colab GPU.

For more information see Modifying StarGAN V2 using Modulated Convolutions.

Images generation

For images generation:

1. Unpack pretrained weights in repository root.
2. Run generate_images.ipynb file.

Training the model

For images generation see train_colab.ipynb file.

Requirements

• python;
• pytorch;
• torchvision;
• opencv;
• munch;
• kornia;
• ffmpeg.