Towards Robust Point Cloud Models with Context-Consistency Network and Adaptive Augmentation

Guanlin Li, Guowen Xu, Han Qiu, Ruan He, Jiwei Li, Tianwei Zhang

Abstract: 3D point cloud models based on deep neural networks were proven to be vulnerable to adversarial examples, with a quantity of novel attack techniques proposed by researchers recently. It is of paramount importance to preserve the robustness of 3D models under adversarial environments, considering their broad application in safety- and security-critical tasks. Unfortunately, defenses for 3D models are much less studied compared to 2D image models. In this paper, we reason about the vulnerability of 3D models based on the mutual information theory. Furthermore, we design an effective defense methodology, consisting of two innovations. (1) We introduce CCDGN, a novel 3D DNN architecture which includes robust and light-weight modules to alleviate adversarial examples. (2) We propose AA-AMS a novel data augmentation strategy to adaptively balance the model usability and robustness. Extensive evaluations indicate the integration of the two techniques provides much more robustness than existing defense solutions for 3D models.