/FaceID-GAN

this is a re-implementation of CVPR2018 paper "FaceID-GAN"

Primary LanguagePython

FaceID-GAN

this is a re-implementation of CVPR2018 paper "FaceID-GAN" using Pytorch

note: I don't finish this work because model is so big that my compute doesn't support training, but I haved done most of work such as training a P net according to paper and building model arctiture. Maybe you need to optimize my code, because I don't test it.

dependencies

  • Pytorch >=0.4
  • python3
  • pip install face-alignment

get pre-trained P net

  • firstly to get 68 key points of one face, using this great work
  • using open source code to get 3DMM parameters for training P net
  • change "./train_p/coonfig.py" and "./train_p/vector_loader.py" to adapt to your environment.
  • run "./train_p/train_pnet.py", and p outputs 235-dims vector whose formulation is [Pose_Para;Shape_Para;Exp_Para]

notes that I have trained p net, you can download from here. Code is "1yvm".

acknowledgement

  • thanks for first author of FaceID-GAN to reply my email and untie my confusion.
  • thanks for the wonderful open source works to develop this project,such as BEGAN,face_alignment.