Transferable and Undefendable Point Cloud Attacks via Medial Axis Transform

2025-07-24

AI Summary

Point cloud adversarial attacks are critical for evaluating the robustness of 3D models, yet existing white-box methods suffer from poor transferability and vulnerability to defensive mechanisms. To address this, we propose the first adversarial attack framework leveraging the Medial Axis Transform (MAT), introducing structure-level perturbations in the MAT representation space. Our method employs an autoencoder to learn compact, geometrically meaningful MAT embeddings, optimizes adversarial perturbations directly in the medial axis space, and incorporates dropout regularization to prevent perturbation collapse. By operating on the intrinsic skeletal structure rather than raw point coordinates, our approach mitigates sensitivity to geometric redundancy and noise inherent in point clouds. Consequently, it achieves significantly improved cross-model transferability and resilience against prevalent defenses—including input transformation and randomization. Evaluated across multiple state-of-the-art 3D classifiers and defense configurations, our method attains an average 12.7% gain in transfer attack success rate and consistently outperforms existing approaches in adversarial robustness.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Multiagent Systems: Adversarial Agents

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Studying adversarial attacks on point clouds is essential for evaluating and improving the robustness of 3D deep learning models. However, most existing attack methods are developed under ideal white-box settings and often suffer from limited transferability to unseen models and insufficient robustness against common defense mechanisms. In this paper, we propose MAT-Adv, a novel adversarial attack framework that enhances both transferability and undefendability by explicitly perturbing the medial axis transform (MAT) representations, in order to induce inherent adversarialness in the resulting point clouds. Specifically, we employ an autoencoder to project input point clouds into compact MAT representations that capture the intrinsic geometric structure of point clouds. By perturbing these intrinsic representations, MAT-Adv introduces structural-level adversarial characteristics that remain effective across diverse models and defense strategies. To mitigate overfitting and prevent perturbation collapse, we incorporate a dropout strategy into the optimization of MAT perturbations, further improving transferability and undefendability. Extensive experiments demonstrate that MAT-Adv significantly outperforms existing state-of-the-art methods in both transferability and undefendability. Codes will be made public upon paper acceptance.

Problem

• Enhancing adversarial attack transferability on point clouds
• Improving robustness against common defense mechanisms
• Perturbing medial axis transform for structural-level adversarialness

Innovation

• Perturbs medial axis transform for adversarial attacks
• Uses autoencoder for compact geometric representation
• Incorporates dropout to enhance transferability robustness