Adversarial-CAE is a good framework for face synthesize and data arguement. It has been originally introduced in this research GCNarticle & Pix2Pix article This repository contains a tensorflow implementation of Adversarial-CAE.

• tensorflow
• openface

We used two kinds of datasets, self dataset and KDEF For the first time to run, you should prepare data first, run command:

python main.py \
	--generate_data \
	--dataset_dir DATASET_DIR \

It will generate data from dataset directory, and if you want to use extra data/two dataset

python main.py \
	--generate_data \
	--use_extra_data \
	--dataset_dir DATASET_DIR \
	--extra_data_dir EXTRA_DATA_DIR

If your self dataset's quality is not so good, you can perform face alignment to your own dataset, just add params

--face_align

When second run main.py, maybe data file is exist, please remove param

--generate_data && --dataset_dir DATASET_DIR 

At the test/sample time, you just need run

python main.py --sampling