zhangqianhui/GazeAnimation

GazeAnimation - Official Tensorflow Implementation

Dual In-painting Model for Unsupervised Gaze Correction and Animation in the Wild
Jichao Zhang, Jingjing Chen, Hao Tang, Wei Wang, Yan Yan, Enver Sangineto, Nicu Sebe
In ACM MM 2020.

Paper: https://arxiv.org/abs/2008.03834

Network Architecture

Dependencies

Python=3.6
pip install -r requirements.txt

Or Using Conda

-conda create -name GazeA python=3.6
-conda install tensorflow-gpu=1.9 or higher

Other packages installed by pip.

Usage

• Clone this repo:

git clone https://github.com/zhangqianhui/GazeAnimation.git
cd GazeAnimation

• Download the CelebAGaze dataset

  Download the tar of CelebAGaze dataset from Google Driver Linking.

  cd your_path
  tar -xvf CelebAGaze.tar

  Please edit the options.py and change your dataset path

• VGG-16 pretrained weights

wget http://download.tensorflow.org/models/vgg_16_2016_08_28.tar.gz .
tar -xvf vgg_16_2016_08_28.tar.gz

Please edit the options.py and change your vgg path

• Pretrained model for PAM module.

Download it from PAM Pretrained model. PLease unzip it in pam_dir and don't contain the sub-dir.

• Train the model using command line with python

python train.py --use_sp --gpu_id='0' --exper_name='log8_7' --crop_w=50 --crop_h=30

• Test the model

python test.py --exper_name='log8_7' --gpu_id='0' --crop_h=30 --crop_w=50 --test_sample_dir='test_sample_dir' --checkpoints='checkpoints'

Or Using scripts for training

bash scripts/train_log8_7.sh

Using scripts for testing and pretained model can be downloaded [V1]
[V2]. Unzip pretrained.zip and move files into 'experiments/checkpoints'

bash scripts/test_log8_7.sh

Experiment Result

Gaze Correction

Gaze Animation

Citation

@inproceedings{zhangGazeAnimation,
  title={Dual In-painting Model for Unsupervised Gaze Correction and Animation in the Wild},
  author={Jichao Zhang, Jingjing Chen, Hao Tang, Wei Wang, Yan Yan, Enver Sangineto, Nicu Sebe},
  booktitle={ACM MM},
  year={2020}
}