Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Keke Tang, Weiyao Ke, Weilong Peng, Xiaofei Wang, Ziyong Du, Zhize Wu, Peican Zhu, Zhihong Tian

TL;DR

This work tackles the problem of imperceptible adversarial attacks on 3D point clouds by identifying surface deviation as the key source of perceptibility. It introduces a point-to-surface (P2S) field, implemented as ${\mathcal{F}(q) = \nabla_q \log Q_S(q)}$, learned via a denoising network to drag perturbed points back to the underlying surface and a distance-aware magnitude to guide perturbations. The authors integrate the P2S field into an iterative attack framework, adjusting perturbation directions and magnitudes to produce highly imperceptible adversarial point clouds while maintaining strong attack performance. Extensive experiments across multiple models and datasets demonstrate superior imperceptibility compared to state-of-the-art methods and show the approach generalizes to other iterative attacks, highlighting its impact for robust evaluation of 3D DNN robustness.

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.

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

TL;DR

This work tackles the problem of imperceptible adversarial attacks on 3D point clouds by identifying surface deviation as the key source of perceptibility. It introduces a point-to-surface (P2S) field, implemented as , learned via a denoising network to drag perturbed points back to the underlying surface and a distance-aware magnitude to guide perturbations. The authors integrate the P2S field into an iterative attack framework, adjusting perturbation directions and magnitudes to produce highly imperceptible adversarial point clouds while maintaining strong attack performance. Extensive experiments across multiple models and datasets demonstrate superior imperceptibility compared to state-of-the-art methods and show the approach generalizes to other iterative attacks, highlighting its impact for robust evaluation of 3D DNN robustness.

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.

Paper Structure

This paper contains 10 sections, 7 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Formulation
  4. Method
  5. DNN-based Point-to-Surface Field
  6. P2S Field-guided Imperceptible Adversarial Attacks
  7. Experimental Results
  8. Experimental Setup
  9. Performance Comparison and Analysis
  10. Conclusion

Figures (3)

  • Figure 1: Illustration of our point-to-surface (P2S) field-guided adversarial attacks. For each adversarial point in a point cloud, we adjust its direction using the P2S field to bring it closer to the surface, making the perturbation imperceptible.
  • Figure 2: Visualizations of original and adversarial point clouds generated to fool PointNet on ModelNet40 by various adversarial attack methods. The ground truth and predicted labels are marked in blue and gray below the images.
  • Figure 3: Visualization of adversarial point clouds generated by various attack methods in attacking PointNet, with and without guidance from the P2S field. The ground truth and predicted labels are marked in blue and gray below the images.