/C-GAN

Primary LanguageC++

C-GAN

Make realistic face images from 3D simulated

Instructions:

Install:

  • keras
  • tensorflow-gpu
  • scipy 1.1.0
  • scikit-image
sudo apt-add-repository ppa:zarquon42/meshlab
sudo apt-get update
sudo apt-get install meshlab=1.3.2+dfsg1-2build4

pip install git+https://www.github.com/keras-team/keras-contrib.git

Generating 3D simulated images from still images:

  • Put the still images in "./face3d/input/", while each identity is in a seperate folder.
  • Run:
pyhton face3d.py
  • 3D rendered results will be in "face3d/output"

Using CGAN to refine the 3D simulated images:

Put the train and test 3D simulated data in:

./data/chokepoint/train_sim
./data/chokepoint/train_sim

Put the train and test target data in:

data/chokepoint/train_target
data/chokepoint/train_target

Put the test and train labels (only 3D simulated data needs label) in:

./data/chokepoint/test_sim_labels.txt ./data/chokepoint/train_sim_labels.txt


Run:

python cgan.py


Results will be in "./output"