Transferable 3D Adversarial Shape Completion using Diffusion Models ECCV 2024

The code repository for our paper Transferable 3D Adversarial Shape Completion using Diffusion Models arXiv

Installation

This repository is based on the official code from PVD.

1. Set up environments for the codes. Details please refer to the original Github code.

python==3.6
pytorch==1.4.0
torchvision==0.5.0
cudatoolkit==10.1
matplotlib==2.2.5
tqdm==4.32.1
open3d==0.9.0
trimesh=3.7.12
scipy==1.5.1

2. Download ShapeNet Completion Dataset (https://github.com/xiumingzhang/GenRe-ShapeHD).

3. Download the PVD model checkpoint (https://drive.google.com/drive/folders/1Q7aSaTr6lqmo8qx80nIm1j28mOHAHGiM?usp=sharing).

4. Download the ShapeNet pre-trained 3D point cloud classifiers (https://drive.google.com/file/d/1fgz4OPdPYnb6n5yyhGOjh-N2kKMThVQ4/view?usp=sharing). You can train the target models by following https://github.com/qiufan319/benchmark_pc_attack/tree/master/baselines.

Usage

Simply run completion-attack_uncertain.py to conduct 3D point cloud attacks. Make sure to set the directory accordingly in the parameters.

python completion-attack_uncertain.py 

Reference

Please cite our paper if you found any helpful information:

@article{dai2024transferable,
title={Transferable 3D Adversarial Shape Completion using Diffusion Models},
author={Dai, Xuelong and Xiao, Bin},
journal={arXiv preprint arXiv:2407.10077},
year={2024}
}