README.md

The project is implemented from the paper "On Adversarial Robustness of Point Cloud Semantic Segmentation".

Start

Downloading S3DIS from https://shapenet.cs.stanford.edu/media/indoor3d_sem_seg_hdf5_data.zip by python is not available due to the expired certificate. So either try to use wget https://shapenet.cs.stanford.edu/media/indoor3d_sem_seg_hdf5_data.zip --no-check-certificate or download it manually and then move the files to examples/sem_seg_dense/data/deepgcn/S3DIS.

cd examples/sem_seg_dense
python test.py --attack <attack method>

Attack methods can be changed by chaning the attack method.

Requirements

• Pytorch>=1.4.0
• pytorch_geometric>=1.3.0

Currently, pytorch_geometric does not provide torch 1.7.1. But we can use torch 1.7.0 version of pytorch_geometric. Available versions of pytorch_geometric can be checked in the link

Code Architecture

.
├── utils                   # Common useful modules
├── gcn_lib                 # gcn library
│   ├── dense               # gcn library for dense data (B x C x N x 1)
│   └── sparse              # gcn library for sparse data (N x C)
├── sem_seg_dense           # code for point clouds semantic segmentation on S3DIS

Great thanks for the Deep GCNs project. The attack code is implemented based on the adversarial-attacks-pytorch.