Robust Unsupervised Domain Adaptation for 3D Point Cloud Segmentation Under Source Adversarial Attacks

2025-04-02

AI Summary

Unsupervised domain adaptation (UDA) for 3D point cloud segmentation lacks robustness against adversarial attacks on the source domain. Method: This work introduces, for the first time, a stealthy surface perturbation attack paradigm targeting source-domain point clouds, yielding the synthetically contaminated dataset AdvSynLiDAR. We further propose the Robust Adaptive Framework (AAF), integrating a Robust Long-Tail (RLT) loss, a Keypoint-Sensitive (KPS) extension mechanism, and a dual-decoder architecture to jointly enhance discriminative robustness for long-tail classes and geometric structure reconstruction. Contribution/Results: Evaluated on AdvSynLiDAR, AAF significantly mitigates UDA performance degradation under source-domain adversarial perturbations: overall segmentation mIoU improves by 12.7%, and accuracy on long-tail classes increases by over 23%, demonstrating strong robustness against such attacks.

Technology Category

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

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Unsupervised domain adaptation (UDA) frameworks have shown good generalization capabilities for 3D point cloud semantic segmentation models on clean data. However, existing works overlook adversarial robustness when the source domain itself is compromised. To comprehensively explore the robustness of the UDA frameworks, we first design a stealthy adversarial point cloud generation attack that can significantly contaminate datasets with only minor perturbations to the point cloud surface. Based on that, we propose a novel dataset, AdvSynLiDAR, comprising synthesized contaminated LiDAR point clouds. With the generated corrupted data, we further develop the Adversarial Adaptation Framework (AAF) as the countermeasure. Specifically, by extending the key point sensitive (KPS) loss towards the Robust Long-Tail loss (RLT loss) and utilizing a decoder branch, our approach enables the model to focus on long-tail classes during the pre-training phase and leverages high-confidence decoded point cloud information to restore point cloud structures during the adaptation phase. We evaluated our AAF method on the AdvSynLiDAR dataset, where the results demonstrate that our AAF method can mitigate performance degradation under source adversarial perturbations for UDA in the 3D point cloud segmentation application.

Problem

• Enhancing adversarial robustness in 3D point cloud UDA
• Mitigating performance drop from source domain attacks
• Improving segmentation accuracy for long-tail classes

Innovation

• Stealthy adversarial point cloud generation attack
• AdvSynLiDAR dataset with contaminated LiDAR
• Adversarial Adaptation Framework with RLT loss