PointMLP/README.md

# Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework （ICLR 2022）



[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/rethinking-network-design-and-local-geometry/3d-point-cloud-classification-on-modelnet40)](https://paperswithcode.com/sota/3d-point-cloud-classification-on-modelnet40?p=rethinking-network-design-and-local-geometry)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/rethinking-network-design-and-local-geometry/3d-point-cloud-classification-on-scanobjectnn)](https://paperswithcode.com/sota/3d-point-cloud-classification-on-scanobjectnn?p=rethinking-network-design-and-local-geometry)


<div align="left">
    <a><img src="images/smile.png"  height="70px" ></a>
    <a><img src="images/neu.png"  height="70px" ></a>
    <a><img src="images/columbia.png"  height="70px" ></a>
</div>

[Project Sites]() | [arXiv](https://arxiv.org/abs/2202.07123) | Primary contact: [Xu Ma](mailto:ma.xu1@northeastern.edu)

<div align="center">
  <img src="images/overview.png" width="650px" height="300px">
</div>

Overview of one stage in PointMLP. Given an input point cloud, PointMLP progressively extract local features using residual point MLP blocks. In each stage, we first transform local point using a geometric affine module, then local points are are extracted before and after aggregation respectively. By repeating multiple stages, PointMLP progressively enlarge the receptive field and model entire point cloud geometric information.


## BibTeX

    @inproceedings{
        ma2022rethinking,
        title={Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual {MLP} Framework},
        author={Xu Ma and Can Qin and Haoxuan You and Haoxi Ran and Yun Fu},
        booktitle={International Conference on Learning Representations},
        year={2022},
        url={https://openreview.net/forum?id=3Pbra-_u76D}
    }

## Model Zoo
- The codes/models/logs for submission version (without bug fixed) can be found here [commit:d2b8dbaa](http://github.com/13952522076/pointMLP-pytorch/tree/d2b8dbaa06eb6176b222dcf2ad248f8438582026).

- On ModelNet40, fixed pointMLP achieves a result of **91.5% mAcc** and **94.1% OA ** without voting, logs and pretrained models can be found [[here]](https://web.northeastern.edu/smilelab/xuma/pointMLP/checkpoints/fixstd/modelnet40/pointMLP-20220209053148-404/).
- On ScanObjectNN, fixed pointMLP achieves a result of **84.4% mAcc** and **86.1% OA ** without voting, logs and pretrained models can be found [[here]](https://web.northeastern.edu/smilelab/xuma/pointMLP/checkpoints/fixstd/scanobjectnn/pointMLP-20220204021453/).
- Stay tuned. More elite versions and voting results will be uploaded.



## News & Updates:

- [ ] updated more pretrained models
- [ ] double check the part seg utils
- [ ] project page
- [x] update std bug (unstable testing in previous version)
- [x] paper/codes release

:point_right::point_right::point_right:**NOTE:** The codes/models/logs for submission version (without bug fixed) can be found here [commit:d2b8dbaa](http://github.com/13952522076/pointMLP-pytorch/tree/d2b8dbaa06eb6176b222dcf2ad248f8438582026).




## Install

```bash
# 1. clone this repo
git clone https://github.com/ma-xu/pointMLP-pytorch.git
cd pointMLP-pytorch

# 2. create a conda virtual environment and activate it
conda create -n pointmlp python=3.7 -y
conda activate pointmlp

# 3. install required libs, pytorch 1.8.1, torchvision 0.9.1, etc.
pip install -r requirements.txt

# 4. install CUDA kernels
pip install pointnet2_ops_lib/.
```


## Useage

### Classification ModelNet40
**Train**: The dataset will be automatically downloaded, run following command to train.

By default, it will create a fold named "checkpoints/{modelName}-{msg}-{randomseed}", which includes args.txt, best_checkpoint.pth, last_checkpoint.pth, log.txt, out.txt.
```bash
cd pointMLP-pytorch/classification_ModelNet40
# train pointMLP
python main.py --model pointMLP
# train pointMLP-elite
python main.py --model pointMLPElite
# please add other paramemters as you wish.
```


To conduct voting testing, run
```bash
# please modify the msg accrodingly
python voting.py --model pointMLP --msg demo
```


### Classification ScanObjectNN

- Make data folder and download the dataset
```bash
cd pointMLP-pytorch/classification_ScanObjectNN
mkdir data
cd data
wget http://103.24.77.34/scanobjectnn/h5_files.zip
unzip h5_files.zip
```

- Train pointMLP/pointMLPElite 
```bash
# train pointMLP
python main.py --model pointMLP
# train pointMLP-elite
python main.py --model pointMLPElite
# please add other paramemters as you wish.
```
By default, it will create a fold named "checkpoints/{modelName}-{msg}-{randomseed}", which includes args.txt, best_checkpoint.pth, last_checkpoint.pth, log.txt, out.txt.


### Part segmentation

- Make data folder and download the dataset
```bash
cd pointMLP-pytorch/part_segmentation
mkdir data
cd data
wget https://shapenet.cs.stanford.edu/media/shapenetcore_partanno_segmentation_benchmark_v0_normal.zip --no-check-certificate
unzip shapenetcore_partanno_segmentation_benchmark_v0_normal.zip
```

- Train pointMLP
```bash
# train pointMLP
python main.py --model pointMLP
# please add other paramemters as you wish.
```


## Acknowledgment

Our implementation is mainly based on the following codebases. We gratefully thank the authors for their wonderful works.

[CurveNet](https://github.com/tiangexiang/CurveNet),
[PAConv](https://github.com/CVMI-Lab/PAConv),
[GDANet](https://github.com/mutianxu/GDANet),
[Pointnet2_PyTorch](https://github.com/erikwijmans/Pointnet2_PyTorch)

## LICENSE
PointMLP is under the Apache-2.0 license. 
Please contact the authors for commercial use.