Awesome

Robust Tracking against Adversarial Attacks

:herb: Robust Tracking against Adversarial Attacks

Shuai Jia, Chao Ma, Yibing Song and Xiaokang Yang

European Conference on Computer Vision (ECCV), 2020

Introduction

<img src="https://github.com/joshuajss/RTAA/blob/master/demo/visualization.png" width='700'/><br/>

Deep convolutional neural networks (CNNs) are vulnerable to adversarial attacks.

• We propose to generate adversarial examples to deteriorate the performance for visual object tracking.
• Conversely, we propose to defend deep trackers against adversarial attacks that eliminate their effect to alleviate performance drops caused by the adversarial attack.
• We choose two typical trackers, DaSiamRPN and RT-MDnet.

Prerequisites

The environment follows the tracker you intend to attack：

• The specific setting and pretrained model for DaSiamPRN can refer to Code_DaSiamRPN

python = 2.7, pytorch = 1.2.0, opencv-python = 4.2.0.32

• The specific setting and pretrained model for RT-MDNet can refer to Code_RT-MDNet

Results

Result for DaSiamRPN on multiple datasets

Result for RT-MDNet on multiple datasets

:herb: All raw results are available. [Google_drive] [Baidu_Disk] Code: 5ex9

Quick Start

:herb: The code of adversarial attack on DaSiamRPN is released!!

• You should download the OTB2015 dataset in data folder.
• Please download the pretrained model in Code_DaSiamRPN.

Test the original performance on OTB2015 dataset, please using the follwing command.

cd DaSiamRPN/code
python test_otb.py

Test the adversarial attack performance on OTB2015 dataset, please using the follwing command.

cd DaSiamRPN/code
python test_otb_attack.py

Test the adversarial defense performance on OTB2015 dataset, please using the follwing command.

cd DaSiamRPN/code
python test_otb_defense.py

-v can be used to visualize the tracking results.

Demo

<img src="https://github.com/joshuajss/RTAA/blob/master/demo/attack_otb100.gif" width='300'/> <img src="https://github.com/joshuajss/RTAA/blob/master/demo/defense_otb100.gif" width='300'/><br/>         <img src="https://github.com/joshuajss/RTAA/blob/master/demo/legend.png" width='400'/><br/>

Citation

If any part of our paper and code is helpful to your work, please generously citing:

@inproceedings{jia-eccv20-RTAA,
  title={Robust Tracking against Adversarial Attacks},
  author={Jia, Shuai and Ma, Chao and Song, Yibing and Yang, Xiaokang},
  booktitle={European Conference on Computer Vision},
  year={2020}
}

@inproceedings{zhu-eccv18-dasiamrpn,
  title={Distractor-aware Siamese Networks for Visual Object Tracking},
  author={Zhu, Zheng and Wang, Qiang and Li, Bo and Wu, Wei and Yan, Junjie and Hu, Weiming},
  booktitle={European Conference on Computer Vision},
  year={2018}
}

@InProceedings{jung-eccv19-rtmdnet,
author = {Jung, Ilchae and Son, Jeany and Baek, Mooyeol and Han, Bohyung},
title = {Real-Time MDNet},
booktitle = {European Conference on Computer Vision (ECCV)},
month = {Sept},
year = {2018}
}

Thank you!

License

Licensed under an MIT license.