GANimation-tf

A tensorflow implementation of GANimation

paper: https://arxiv.org/abs/1807.09251

Author's implementation: https://github.com/albertpumarola/GANimation

Requirements

• ubuntu 14.04
• python 3.6
• opencv 3.4.3
• tensorflow-gpu 1.12
• face-recognition 1.2.3

Data

1. Apply for Emotionet dataset and select 200k images and save at folder 'data/emotionet'.
2. Extract AU feature by OpenFace and save at folder 'data/aus_openface'.

FeatureExtraction -fdir src_image_dir/ -aus

3. Detect and crop face by face-recognition library and resize to 128x128x3 then save at 'data/imgs' (use 'data/face_crop.py').
4. Generate 'aus.pkl' (please refer to paper author's github) at 'data/' (use 'data/pkl_generate.py').

Training

python train.py

Testing

python test.py