https://github.com/skyworkai/pointcloudmamba

Point Cloud Mamba: Point Cloud Learning via State Space Model
https://github.com/skyworkai/pointcloudmamba

Last synced: 6 days ago
JSON representation

Point Cloud Mamba: Point Cloud Learning via State Space Model

• Host: GitHub
• URL: https://github.com/skyworkai/pointcloudmamba
• Owner: SkyworkAI
• Created: 2024-03-25T05:25:41.000Z (8 months ago)
• Default Branch: main
• Last Pushed: 2024-06-03T13:51:55.000Z (6 months ago)
• Last Synced: 2024-10-28T08:41:41.948Z (23 days ago)
• Language: Python
• Homepage:
• Size: 854 KB
• Stars: 75
• Watchers: 6
• Forks: 7
• Open Issues: 9
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        


# [Point Cloud Mamba: Point Cloud Learning via State Space Model](https://arxiv.org/abs/2403.00762)

Tao Zhang, Xiangtai Li*, Haobo Yuan, Shunping Ji, Shuicheng Yan





## News

- All codes and weights are available.

## Features

- PCM introduces Mamba to point cloud analysis.

- PCM can perform global modeling while maintaining linear computational complexity.

- PCM outperforms PointNeXt on the ScanObjectNN, ModelNet40, and ShapeNetPart datasets.

## Install 

See [Installation Instructions](INSTALL.md).

## Getting Started

See [Preparing Datasets for PCM](data/README.md).

See [Getting Started with PCM](GETTING_STARTED.md).

## Demos

### ShapeNetPart



## Performance

### 3-D Point Cloud Classification



### 3-D Point Cloud Segmentation



```BibTeX

@article{zhang2024point,

  title={Point Cloud Mamba: Point Cloud Learning via State Space Model},

  author={Zhang, Tao and Li, Xiangtai and Yuan, Haobo and Ji, Shunping and Yan, Shuicheng},

  journal={arXiv preprint arXiv:2403.00762},

  year={2024}

}

```

## Acknowledgement

This repo is based on [PointNeXt](https://github.com/guochengqian/PointNeXt), 

[PointMLP](https://github.com/ma-xu/pointMLP-pytorch), [Mamba](https://github.com/state-spaces/mamba),

[Vim](https://github.com/hustvl/Vim), and [Pontcept](https://github.com/Pointcept/Pointcept).

Thanks for their excellent works.