Geometry-Aware Generation of Adversarial Point Clouds

Wen, Yuxin; Lin, Jiehong; Chen, Ke; Chen, C. L. Philip; Jia, Kui

doi:10.1109/tpami.2020.3044712

Cited by 73 publications

(68 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…3D adversarial attacks aim to generate 3D adversarial samples in a human-unnoticeable way. 3D adversarial samples usually consist of two types: adversarial point cloud (Xiang et al, 2019;Tsai et al, 2020) and adversarial mesh (Wen et al, 2020;Zhang et al, 2021a). Currently, most 3D adversarial attacks are about point cloud.…”

Section: D Adversarial Attacksmentioning

confidence: 99%

“…However, those adversarial point clouds usually contain a lot of outliers, which are not human-unnoticeable. To solve this problem, the following works (Wen et al, 2020;Tsai et al, 2020) focus on generating adversarial point cloud with much less outliers. Tsai et al (2020) proposed kNN attack that aims to generating smooth perturbation by adding chamfer distance and kNN distance to loss function as the regularization terms during optimization.…”

Section: D Adversarial Attacksmentioning

confidence: 99%

“…Tsai et al (2020) proposed kNN attack that aims to generating smooth perturbation by adding chamfer distance and kNN distance to loss function as the regularization terms during optimization. Wen et al (2020) proposed the GeoA 3 attack which crafts adversarial point cloud in a geometric-aware way and is more human imperceptible.…”

Section: D Adversarial Attacksmentioning

confidence: 99%

“…The adversarial samples mislead the deep learning models by introducing small perturbation to the clean data. In the case of point cloud data, researchers mostly perform adversarial attacks on point cloud data by adding, deleting some points, or changing their coordinates (Tsai et al, 2020;Wen et al, 2020;Xiang et al, 2019;Zhou et al, 2020;Hamdi et al, 2020;Ma et al, 2020).…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Boosting 3D Adversarial Attacks with Attacking On Frequency

Liu¹,

Zhang²,

Chen³

et al. 2022

Preprint

View full text Add to dashboard Cite

Deep neural networks (DNNs) have been shown to be vulnerable to adversarial attacks. Recently, 3D adversarial attacks, especially adversarial attacks on point clouds, have elicited mounting interest. However, adversarial point clouds obtained by previous methods show weak transferability and are easy to defend. To address these problems, in this paper we propose a novel point cloud attack (dubbed AOF) that pays more attention on the low-frequency component of point clouds. We combine the losses from point cloud and its lowfrequency component to craft adversarial samples. Extensive experiments validate that AOF can improve the transferability significantly compared to state-of-the-art (SOTA) attacks, and is more robust to SOTA 3D defense methods. Otherwise, compared to clean point clouds, adversarial point clouds obtained by AOF contain more deformation than outlier.

show abstract

Section: D Adversarial Attacksmentioning

confidence: 99%

Section: D Adversarial Attacksmentioning

confidence: 99%

Section: D Adversarial Attacksmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Boosting 3D Adversarial Attacks with Attacking On Frequency

Liu¹,

Zhang²,

Chen³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…We are interested in understanding if the use of an additional input prevents adversarial attacks on the other input. Though we could utilize adversarial attacks on the LIDAR input like previous work [11,2,12,13], we instead choose to focus on modifying the image input. This is because we find that the model relies more heavily on LIDAR data and successful attacks using modification of just the LIDAR is more trivial.…”

Section: Introductionmentioning

confidence: 99%

Sensor Adversarial Traits: Analyzing Robustness of 3D Object Detection Sensor Fusion Models

Park

Liu

Chen

et al. 2021

2021 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

A critical aspect of autonomous vehicles (AVs) is the object detection stage, which is increasingly being performed with sensor fusion models: multimodal 3D object detection models which utilize both 2D RGB image data and 3D data from a LIDAR sensor as inputs. In this work, we perform the first study to analyze the robustness of a high-performance, open source sensor fusion model architecture towards adversarial attacks and challenge the popular belief that the use of additional sensors automatically mitigate the risk of adversarial attacks. We find that despite the use of a LIDAR sensor, the model is vulnerable to our purposefully crafted image-based adversarial attacks including disappearance, universal patch, and spoofing. After identifying the underlying reason, we explore some potential defenses and provide some recommendations for improved sensor fusion models.

show abstract