TY - GEN
T1 - 3DeformRS
T2 - 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022
AU - Gabriel Perez, S.
AU - Perez, Juan C.
AU - Alfarra, Motasem
AU - Giancola, Silvio
AU - Ghanem, Bernard
N1 - Funding Information:
Acknowledgements. This publication is based upon work supported by the King Abdullah University of Science and Technology (KAUST) Office of Sponsored Research (OSR) under Award No. OSR-CRG2019-4033. We would also like to thank Jesús Zarzar for the help and discussions.
Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - 3D computer vision models are commonly used in security-critical applications such as autonomous driving and surgical robotics. Emerging concerns over the robustness of these models against real-world deformations must be addressed practically and reliably. In this work, we propose 3DeformRS, a method to certify the robustness of point cloud Deep Neural Networks (DNNs) against real-world deformations. We developed 3DeformRS by building upon recent work that generalized Randomized Smoothing (RS) from pixel-intensity perturbations to vector-field deformations. In particular, we specialized RS to certify DNNs against parameterized deformations (e.g. rotation, twisting), while enjoying practical computational costs. We leverage the virtues of 3DeformRS to conduct a comprehensive empirical study on the certified robustness of four representative point cloud DNNs on two datasets and against seven different deformations. Compared to previous approaches for certifying point cloud DNNs, 3DeformRS is fast, scales well with point cloud size, and provides comparable-to-better certificates. For instance, when certifying a plain PointNet against a 3° z-rotation on 1024-point clouds, 3DeformRS grants a certificate 3× larger and 20× faster than previous work 11Code:https://github.com/gaperezsa/3DeformRS.
AB - 3D computer vision models are commonly used in security-critical applications such as autonomous driving and surgical robotics. Emerging concerns over the robustness of these models against real-world deformations must be addressed practically and reliably. In this work, we propose 3DeformRS, a method to certify the robustness of point cloud Deep Neural Networks (DNNs) against real-world deformations. We developed 3DeformRS by building upon recent work that generalized Randomized Smoothing (RS) from pixel-intensity perturbations to vector-field deformations. In particular, we specialized RS to certify DNNs against parameterized deformations (e.g. rotation, twisting), while enjoying practical computational costs. We leverage the virtues of 3DeformRS to conduct a comprehensive empirical study on the certified robustness of four representative point cloud DNNs on two datasets and against seven different deformations. Compared to previous approaches for certifying point cloud DNNs, 3DeformRS is fast, scales well with point cloud size, and provides comparable-to-better certificates. For instance, when certifying a plain PointNet against a 3° z-rotation on 1024-point clouds, 3DeformRS grants a certificate 3× larger and 20× faster than previous work 11Code:https://github.com/gaperezsa/3DeformRS.
KW - 3D from multi-view and sensors
KW - Adversarial attack and defense
UR - http://www.scopus.com/inward/record.url?scp=85140199096&partnerID=8YFLogxK
U2 - 10.1109/CVPR52688.2022.01474
DO - 10.1109/CVPR52688.2022.01474
M3 - Conference contribution
AN - SCOPUS:85140199096
T3 - Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
SP - 15148
EP - 15158
BT - Proceedings - 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022
PB - IEEE Computer Society
Y2 - 19 June 2022 through 24 June 2022
ER -