Eidos: Efficient, Imperceptible Adversarial 3D Point Clouds

TL;DR

Eidos is presented, a framework providing Efficient Imperceptible aDversarial attacks on 3D pOint cloudS, thereby enabling a runtime-imperceptibility trade-off and showing Eidos' superiority with respect to efficiency as well as imperceptibility.

Abstract

Classification of 3D point clouds is a challenging machine learning (ML) task with important real-world applications in a spectrum from autonomous driving and robot-assisted surgery to earth observation from low orbit. As with other ML tasks, classification models are notoriously brittle in the presence of adversarial attacks. These are rooted in imperceptible changes to inputs with the effect that a seemingly well-trained model ends up misclassifying the input. This paper adds to the understanding of adversarial attacks by presenting Eidos, a framework providing Efficient Imperceptible aDversarial attacks on 3D pOint cloudS. Eidos supports a diverse set of imperceptibility metrics. It employs an iterative, two-step procedure to identify optimal adversarial examples, thereby enabling a runtime-imperceptibility trade-off. We provide empirical evidence relative to several popular 3D point cloud classification models and several established 3D attack methods, showing Eidos' superiority with respect to efficiency as well as imperceptibility.

Figures (19)

• Figure 1: Visualization of adversarial point clouds. It shows the original sample and adversarial distortions generated by different attack methods. Eidos here is used with $D_{L_2}$ as imperceptibility regularization term. The number displayed in the bottom right denotes the mean $L_2$ norm of distortions, and it is clear that Eidos results in better imperceptibility than SI-Adv, GeoA-${{ \textbf{3}}}$, and GSDA.
• Figure 2: Operating charateristics on PointNet for our attack constrained by different imperceptibility regularization.
• Figure 3: Visualization of adversarial distortions produced by baseline methods and Eidos.
• Figure 4: Operating characteristics of $P_{suc}$vs. HD and Curv on PointNet with DUP-Net w.r.t. Table \ref{['tab:defense']}.
• Figure A5: Various indicators under different step sizes with imperceptibility regularization of Eidos. Maximum iteration: $100$; Step size: $\epsilon \in [0.001, 0.01, 0.02, \cdots, 0.06, 0.1]$.