cook / FHR_alignment

Code for our AAAI'19 paper "Towards Highly Accurate and Stable Face Alignment for High-Resolution Videos"

Fractional Heatmap Regression

[Paper] [Slides] [Supp][TencentYouTuResearch]

Citation

If you find this work useful in your research, please consider citing (* indicates equal contributions):

@inproceedings{tai-FHR-2019,
  title={Towards Highly Accurate and Stable Face Alignment for High-Resolution Videos},
  author={Tai, Ying* and Liang, Yicong* and Liu, Xiaoming and Duan, Lei and Li, Jilin and Wang, Chengjie and Huang, Feiyue and Chen, Yu},
  booktitle={The AAAI Conference on Artificial Intelligence (AAAI)},
  year={2019}
}

Prerequisites

• Torch
• Linux
• NVIDIA GPU + CUDA CuDNN

Getting Started

Setup

Clone the github repository:

git  clone https://github.com/tyshiwo/FHR_alignment.git
cd FHR_alignment

Data & Models

1. Download the training and testing data include datasetsets 300W and 300VW from [Google drive]. Put it into the root path (i.e., FHR_alignment/data)
2. Download the pretrained models for datasets 300W and 300VW from [Google drive]. Put it into the root path (i.e., FHR_alignment/models)

Training

For dataset 300W:

cd training_code
sh exec_train_300W_fhr.sh

For dataset 300VW:

sh exec_train_300VW_fhr.sh

Testing

For dataset 300W:

cd testing_code
sh test_300W_fhr.sh

For dataset 300VW:

sh test_300VW_fhr.sh