Deep Manifold Attack on Point Clouds via Parameter Plane Stretching

Authors: Keke Tang, Jianpeng Wu, Weilong Peng, Yawen Shi, Peng Song, Zhaoquan Gu, Zhihong Tian, Wenping Wang

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments show that adversarial point clouds generated by manifold attack are smooth, undefendable and transferable, and outperform those samples generated by the state-of-the-art non-manifold ones.
Researcher Affiliation Academia 1 Guangzhou University 2 Singapore University of Technology and Design 3 Harbin Institute of Technology (Shenzhen) 4 Peng Cheng Laboratory 5 Texas A&M University
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide an explicit statement about releasing source code or a link to a code repository for the described methodology.
Open Datasets Yes We adopt two public datasets for evaluation: Shape Net Part (Chang et al. 2015) and Model Net40 (Wu et al. 2015).
Dataset Splits Yes We select 14007 point clouds for training and 2874 for testing on Shape Net Part, while 9843 for training and 2468 for testing on Model Net40 following (Qi et al. 2017b).
Hardware Specification Yes Both the pretrain of manifold auto-encoder and the training of manifold attack framework are performed on a workstation with one NVIDIA RTX 2080Ti GPU for 1000 epochs.
Software Dependencies No The paper mentions using 'Py Torch' but does not specify a version number for it or any other key software dependencies.
Experiment Setup Yes the mapping representation θP is denoted with the codeword in size of 1 1024, and the parameter plane is denoted with a 45 45 point grid in the range of [ 0.3, 0.3]. For TPS transformation, we use 4 4 control points. The offset prediction network Fo is implemented with MLP (3 64 128 1024)-Max Pool-FC (1024 512 256 32)-Tanh to predict the offsets of control points along two axes in the range of [ 1.0, 1.0]. [...] α is a weighting parameter, setting as 0.2 in our paper.