Benchmark of Pointcloud Adversarial Attacks and Defenses

Introduction

Point cloud is an important 3D representation that is widely used in various security-critical applications. Although deep learning on point cloud has exhibited high performance, recent research has shown that deep learning models on point cloud are vulnerable to adversarial attacks. To help researchers more easily compare existing attacks and defenses, in this repo we provide PyTorch implementations for common baseline attacks and defenses. All our code is modified or integrated from other existing code.

Requirement

Our project is developed using Python 3.8, PyTorch 1.11.0 with CUDA11.6. We recommend you to use anaconda for dependency configuration.

First create an anaconda environment called benchmark by Cancel changes

conda create -n benchmark python=3.8

conda activate benchmark

All the required packages are listed in the requirements.txt file. First use pip to install them by

python -m pip install -r requirements.txt

Then, you need to install torch, torchvision and torch-scatter manually to fit in with your server environment (e.g. CUDA version). For the torch and torchvision used in my project, run

conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorchCancel changes

Project Structure

The code in this project is mainly divided into two folders, each of which has a detailed README file explaining its usage. Here I will briefly outline the structure of this repo.
baselines/ contains code for training/testing the victim models as well as some baseline attack and defense methods.
AT/ contains code for adversarial training ConvONet/ contains code for IF-defense pointcutmix/contains code for pointcutmix Please go to each folder to see how to use the code in detail.

Result

Classification accuracy of ModelNet40 under black-box attacks and defense.
baslines: PointNet

Acknowledgements

We thank the authors of following works for opening source their excellent codes.

• PointNet/PointNet++, DGCNN, PointConv
• Perturb/Add attack, kNN attack, Drop attack
• PU-Net, DUP-Net
• ONet
• IF-Defense
• GEOA3