Imperceptible Adversarial Attacks on Point Clouds Guided by Point-to-Surface Field

TL;DR

This paper introduces a novel point-to-surface (P2S) field that guides adversarial attacks on 3D point clouds to be more imperceptible by aligning perturbations with the underlying surface, improving robustness testing.

Contribution

The paper proposes a new P2S field using a denoising network to better align adversarial perturbations with the surface, enhancing imperceptibility in point cloud attacks.

Findings

P2S-guided attacks are more imperceptible than existing methods.

The approach outperforms state-of-the-art adversarial attack techniques.

Enhanced surface-aware perturbation improves attack effectiveness.

Abstract

Adversarial attacks on point clouds are crucial for assessing and improving the adversarial robustness of 3D deep learning models. Traditional solutions strictly limit point displacement during attacks, making it challenging to balance imperceptibility with adversarial effectiveness. In this paper, we attribute the inadequate imperceptibility of adversarial attacks on point clouds to deviations from the underlying surface. To address this, we introduce a novel point-to-surface (P2S) field that adjusts adversarial perturbation directions by dragging points back to their original underlying surface. Specifically, we use a denoising network to learn the gradient field of the logarithmic density function encoding the shape's surface, and apply a distance-aware adjustment to perturbation directions during attacks, thereby enhancing imperceptibility. Extensive experiments show that adversarial attacks guided by our P2S field are more imperceptible, outperforming state-of-the-art methods.