对抗样本(论文解读十四): advPattern Physical World Attacks on Deep Person Re-Identification via Adversarially T

advPattern Physical World Attacks on Deep Person Re-Identification via Adversarially Transformable

Zhibo Wang†, Siyan Zheng†, Mengkai Song†, Qian Wang†,∗, Alireza Rahimpour‡, Hairong Qi‡

†Key Laboratory of Aerospace Information Security and Trusted Computing, Ministry of Education, School of Cyber Science and Engineering, Wuhan University, P. R. China
‡Dept. of Electrical Engineering and Computer Science, University of Tennessee, Knoxville, USA

发表于：ICCV_2019_paper

摘要

人再识别(re-ID)是一项跨摄像头视图匹配人的任务。

我们第一次尝试对深度re-id实施健壮的物理世界攻击。

我们提出了一种新的攻击算法，称为advPattern，用于生成衣服上的对抗性图案，

该算法通过学习不同相机之间图像对的变化，将来自同一相机的图像特征拉近，同时将来自不同相机的特征推远。

数据集：Market1501 and our established PRCS dataset