Deepfake with Synthetic Face Image project

Recently, Deepfake and Generative Adversarial Network (GAN) are two hot topics in Deep Learning filed. In this mini project, this project is to build an application to bridge these two techniques.It uses Deepfake technique to replace the face in destination image with any synthetic human faces generated by the GAN.

To run this project on your local machine, please follow the next steps below:

1. Please manually download the pre-trained pg-GAN model (provided by Nvidia), the trained feature extractor network, and the discovered feature axis from the link https://www.dropbox.com/sh/y1ryg8iq1erfcsr/AAB--PO5qAapwp8ILcgxE2I6a?dl=0

2. Decompress the downloaded files and put it in project directory as the following format

   root(d):
     asset_model(d):
       karras2018iclr-celebahq-1024x1024.pkl   # pretrained GAN from Nvidia
       cnn_face_attr_celeba(d):
         model_20180927_032934.h5              # trained feature extractor network
     asset_results(d):
       pg_gan_celeba_feature_direction_40(d):
         feature_direction_20181002_044444.pkl # feature axes
   

References

TL-GAN: transparent latent-space GAN. SummitKwan. Retrieved from https://github.com/SummitKwan/transparent_latent_gan

FaceSwap. wuhuikai. Retrieved from https://github.com/wuhuikai/FaceSwap