Imperceptible 3D Point Cloud Attacks on Lattice-based Barycentric Coordinates

Authors: Keke Tang, Ziyong Du, Weilong Peng, Xiaofei Wang, Daizong Liu, Ligang Liu, Zhihong Tian

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments validate that integrating these local parametric constraints into conventional adversarial attacks yields superior imperceptibility, outperforming state-of-the-art methods.
Researcher Affiliation Academia 1Guangzhou University 2University of Science and Technology of China 3Peking University
Pseudocode No The paper describes methods and formulations but does not contain a clearly labeled pseudocode or algorithm block.
Open Source Code No The paper does not contain any explicit statements about releasing source code for the described methodology, nor does it provide a link to a code repository.
Open Datasets Yes We evaluate on Model Net40 (Wu et al. 2015) and Shape Net Part (Chang et al. 2015), sampling 1,024 points per cloud following (Xiang, Qi, and Li 2019).
Dataset Splits No The paper mentions "sampling 1,024 points per cloud" and using Model Net40 and Shape Net Part datasets, but does not explicitly provide details about training/test/validation splits (e.g., percentages, sample counts, or specific split files).
Hardware Specification Yes Experiments are conducted on a workstation with dual 2.40 GHz CPUs, 128 GB RAM, and eight NVIDIA RTX 3090 GPUs.
Software Dependencies No We implement the LBC-constrained adversarial attack framework using Py Torch (Paszke et al. 2019). While PyTorch is mentioned, a specific version number for the library itself is not provided.
Experiment Setup Yes Following (Gu et al. 2019), we construct a permutohedral lattice with a resolution of r = 10 along each axis. For lattice subdivision, we compute each cell s density score and mean curvature, normalize them to [0,1], and calculate the subdivision criterion S( ) as their weighted sum with λd = 0.5. The subdivision threshold τ is set so that the top one-third of cells are subdivided. During lattice refinement, we optimize vertex positions over 2,000 iterations, setting λa = 10.0 to maintain cell angles.