Recent work demonstrated the lack of robustness of optical flow networks to physical, patch-based adversarial attacks. The possibility to physically attack a basic component of automotive systems is a reason for serious concerns. In this paper, we analyze the cause of the problem and show that the lack of robustness is rooted in the classical aperture problem of optical flow estimation in combination with bad choices in the details of the network architecture. We show how these mistakes can be rectified in order to make optical flow networks robust to physical, patch-based attacks.
翻译:最近的工作表明光学流动网络缺乏对物理、补丁基对抗性攻击的稳健性,实际攻击汽车系统一个基本组成部分的可能性是引起严重关切的原因。在本文件中,我们分析了问题的原因,并表明缺乏稳健性的根源在于光学流动估计这一古老的孔径问题,同时在网络结构的细节上作出错误的选择。我们展示了如何纠正这些错误,以使光学流动网络能够对物理、补丁基攻击产生强大的影响。