Rethinking Gradient-based Adversarial Attacks on Point Cloud Classification

2025/05/28 by Jun Chen, Xinke Li, Chen, Jun +6
Computer Science · Engineering · #Adversarial Robustness in Machine Learning #3D Shape Modeling and Analysis #Topology Optimization in Engineering

paper · pdf · doi:10.48550/arxiv.2505.21854

Abstract

Gradient-based adversarial attacks are widely used to evaluate the robustness of 3D point cloud classifiers, yet they often rely on uniform update rules that neglect point-wise heterogeneity, leading to perceptible perturbations. We propose two complementary strategies to improve both the effectiveness and imperceptibility of the attack. WAAttack employs weighted gradients to dynamically adjust per-point perturbation magnitudes and uses an adaptive step size strategy to regulate the global perturbation scale. SubAttack partitions the point cloud into subsets and, at each iteration, perturbs only those combinations with high adversarial efficacy and low perceptual saliency. Together, these methods offer a principled refinement of gradient-based attacks for 3D point clouds. Extensive experiments show that our approach consistently outperforms state-of-the-art methods in generating highly imperceptible adversarial examples. The code is available at https://github.com/chenjun0326/WA_SubAttack.

Citations

• Towards a 3D Transfer-based Black-box Attack via Critical Feature Guidance
• NoPain: No-box Point Cloud Attack via Optimal Transport Singular Boundary
• Hide in Thicket: Generating Imperceptible and Rational Adversarial Perturbations on 3D Point Clouds
• 3DHacker: Spectrum-based Decision Boundary Generation for Hard-label 3D Point Cloud Attack
• PointCert: Point Cloud Classification with Deterministic Certified Robustness Guarantees
• Improving transferability of 3D adversarial attacks with scale and shear transformations
• Isometric 3D Adversarial Examples in the Physical World
• Point Cloud Attacks in Graph Spectral Domain: When 3D Geometry Meets Graph Signal Processing
• Shape-invariant 3D Adversarial Point Clouds
• Exploring the Devil in Graph Spectral Domain for 3D Point Cloud Attacks
• Boosting 3D Adversarial Attacks with Attacking On Frequency
• Imperceptible Transfer Attack and Defense on 3D Point Cloud Classification
• Walk in the Cloud: Learning Curves for Point Clouds Shape Analysis
• 3D Adversarial Attacks Beyond Point Cloud
• PAConv: Position Adaptive Convolution with Dynamic Kernel Assembling on Point Clouds
• PCT: Point cloud transformer
• Point-GNN: Graph Neural Network for 3D Object Detection in a Point Cloud
• GraphTER: Unsupervised Learning of Graph Transformation Equivariant Representations via Auto-Encoding Node-wise Transformations
• Revisiting Point Cloud Classification: A New Benchmark Dataset and\n Classification Model on Real-World Data
• KPConv: Flexible and Deformable Convolution for Point Clouds
• A-CNN: Annularly Convolutional Neural Networks on Point Clouds
• Robustness of 3D Deep Learning in an Adversarial Setting
• Extending Adversarial Attacks and Defenses to Deep 3D Point Cloud Classifiers
• PointCloud Saliency Maps
• PointConv: Deep Convolutional Networks on 3D Point Clouds
• Generating 3D Adversarial Point Clouds
• Dynamic Graph CNN for Learning on Point Clouds
• Dynamic Graph CNN for Learning on Point Clouds
• PointCNN: Convolution On X-Transformed Points
• Large-scale Point Cloud Semantic Segmentation with Superpoint Graphs
• Frustum PointNets for 3D Object Detection from RGB-D Data
• VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection
• PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space
• PointNet: Deep Learning on Point Sets for 3D Classification and\n Segmentation
• A Point Set Generation Network for 3D Object Reconstruction from a Single Image
• Multi-View 3D Object Detection Network for Autonomous Driving
• Target-driven Visual Navigation in Indoor Scenes using Deep Reinforcement Learning
• Towards Evaluating the Robustness of Neural Networks

Related

• On Adversarial Robustness of 3D Point Cloud Classification under Adaptive Attacks
• Local Aggressive Adversarial Attacks on 3D Point Cloud
• Generating Adversarial Point Clouds Using Diffusion Model
• PointDP: Diffusion-driven Purification against Adversarial Attacks on 3D Point Cloud Recognition
• Shape-invariant 3D Adversarial Point Clouds