SAAT

A Tensorflow implementation of ICCV2021 paper: Improving Robustness of Facial Landmark Detection by Defending against Adversarial Attacks. PDF:SAAT

Installation Instructions

TensorFlow >= 1.10.1

Python 2.7

Different strategies of perturbation generation

Pretrained models and Masked 300W dataset

Masked 300W: Download

The pre-training models are coming soon.

Databases

./databases:
           /ibug:       
                /image1.jpg     
                 image1.pts       
                 image2.jpg      
                 image2.pts         
           /helen
           /lfpw
           /Masked 300W
        ...  

bbs

./bbs:
     /ibug:
          image1.pts
          image2.pts  
  /helen
  /lfpw
  /Masked 300W
  ...  

Testing

Downlod the pre-trained models and put it into ./ckpt/test, then run Evaluation.py

Bibtex

@InProceedings{Zhu_2021_ICCV,
author    = {Zhu, Congcong and Li, Xiaoqiang and Li, Jide and Dai, Songmin},
title     = {Improving Robustness of Facial Landmark Detection by Defending Against Adversarial Attacks},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
month     = {October},
year      = {2021},
pages     = {11751-11760}}