ModelNet-C (now part of PointCloud-C)

⚠️ This repo has been integrated into our new benchmark PointCloud-C! Please checkout PointCloud-C for any future updates!

Code for the paper "Benchmarking and Analyzing Point Cloud Classification under Corruptions". For the latest updates, see: sites.google.com/view/modelnetc/home

Benchmarking and Analyzing Point Cloud Classification under Corruptions
Jiawei Ren, Liang Pan, Ziwei Liu

arXiv 2022

ModelNet-C [Download Link]

Get Started

Step 0. Clone the Repo

git clone https://github.com/jiawei-ren/ModelNet-C.git
cd ModelNet-C

Step 1. Set Up the Environment

Set up the environment by:

conda create --name modelnetc python=3.7.5
conda activate modelnetc
pip install -r requirements.txt
cd SimpleView/pointnet2_pyt && pip install -e . && cd -
pip install -e modelnetc_utils

Step 2. Prepare Data

Download ModelNet-C by:

cd data
gdown https://drive.google.com/uc?id=1KE6MmXMtfu_mgxg4qLPdEwVD5As8B0rm
unzip modelnet_c.zip && cd ..

Alternatively, you may download ModelNet40-C manually and extract it under data.

Step 3. Download Pretrained Models

Download pretrained models by

gdown https://drive.google.com/uc?id=11RONLZGg0ezxC16n57PiEZouqC5L0b_h
unzip pretrained_models.zip

Alternatively, you may download pretrained models manually and extract it under root directory.

Benchmark on ModelNet-C

Evaluation Commands

Evaluation commands are provided in EVALUATE.md.

Customize ModelNet-C Evaluation for Your Codebase

We have provided evaluation utilities to help you evaluate on ModelNet-C using your own codebase. Please follow CUSTOMIZE.md.

Benchmark Results

Method	Reference	Standalone	mCE	Clean OA
DGCNN	Wang et al.	Yes	1.000	0.926
PointNet	Qi et al.	Yes	1.422	0.907
PointNet++	Qi et al.	Yes	1.072	0.930
RSCNN	Liu et al.	Yes	1.130	0.923
SimpleView	Goyal et al.	Yes	1.047	0.939
GDANet	Xu et al.	Yes	0.892	0.934
CurveNet	Xiang et al.	Yes	0.927	0.938
PAConv	Xu et al.	Yes	1.104	0.936
PCT	Guo et al.	Yes	0.925	0.930
RPC	Ren et al.	Yes	0.863	0.930
DGCNN+PointWOLF	Kim et al.	No	0.814	0.926
DGCNN+RSMix	Lee et al.	No	0.745	0.930
DGCNN+WOLFMix	Ren et al.	No	0.590	0.932
GDANet+WOLFMix	Ren et al.	No	0.571	0.934

*Standalone indicates if the method is a standalone architecture or a combination with augmentation or pretrain.

Todos

• PointMixup
• OcCo
• PointBERT

Cite ModelNet-C

@article{
    ren2022modelnetc,
    title={Benchmarking and Analyzing Point Cloud Classification under Corruptions},
    author={Jiawei Ren and Liang Pan and Ziwei Liu},
    journal={arXiv:2202.03377},
    year={2022},
}

Acknowledgement

This codebase heavily borrows codes from the following repositories:

• SimpleView
• PCT_Pytorch
• GDANet
• CurveNet
• PAConv
• RSMix
• PointWOLF