Björn Browatzki and Chistian Wallraven, Cognitive Systems Lab, Korea University

This is the PyTorch implementation of the ICCV RSLCV-2019 workshop paper "Robust Discrimination and Generation of Faces using Compact, Disentangled Embeddings".

This method is based around two core ideas: first, our framework learns an unsupervised, low-dimensional embedding of faces using an adversarial autoencoder that is able to synthesize high-quality face images. Second, a supervised disentanglement splits the low-dimensional embedding vector into four sub-vectors, each of which contains separated information about one of four major face attributes (pose, identity, expression, and style) that can be used both for discriminative tasks and for manipulating all four attributes in an explicit manner. The resulting architecture achieves state-of-the-art image quality, good discrimination and face retrieval results on each of the four attributes, and supports various face editing tasks using a face representation of only 99 dimensions.

• Linux (not tested on other platforms)
• Python 3.6 or newer
• PyTorch >= 0.4.0
• CUDA >= 9.0
• cuDNN >= 7.1

You may want to setup a virtual environment for this project. See virtualenv or Anaconda.

pip install -r requirements.txt

For training you need:

• VGGFace2
• AffectNet
• VoxCeleb1

For testing you may need (depending on which evaluations you want to run):

• AffectNet
• LFW

Please change the paths in config.py according your dataset locations.

python train.py --sessionname [my_model]

python train.py --sessionname [my_model_dis] -r [my_model] --disent 1 --train_ae 0

python train.py --sessionname [my_model_dis_joint] --r [my_model_dis] --disent 1 --train_ae 1

If you find this project usefull, please cite the following paper:

@inproceedings{browatzki2019,
  title={Robust Discrimination and Generation of Faces using Compact, Disentangled Embeddings },
  author={Browatzki, Bjoern and Wallraven, Christian},
  booktitle={ICCV RSLCV},
  year={2019}
}