# Pointnet_Pointnet2_pytorch

**Repository Path**: ai-models-cn/Pointnet_Pointnet2_pytorch

## Basic Information

- **Project Name**: Pointnet_Pointnet2_pytorch
- **Description**: PointNet 与 PointNet++ 的标准实现，及中文化注释。
- **Primary Language**: Python
- **License**: MIT
- **Default Branch**: cn-pytorch1.8.0
- **Homepage**: https://github.com/yanx27/Pointnet_Pointnet2_pytorch
- **GVP Project**: No

## Statistics

- **Stars**: 2
- **Forks**: 2
- **Created**: 2024-01-19
- **Last Updated**: 2026-01-20

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# PointNet 与 PointNet++ 的 PyTorch 实现

这个仓库是[PointNet](http://openaccess.thecvf.com/content_cvpr_2017/papers/Qi_PointNet_Deep_Learning_CVPR_2017_paper.pdf)
和 [PointNet++](http://papers.nips.cc/paper/7095-pointnet-deep-hierarchical-feature-learning-on-point-sets-in-a-metric-space.pdf) 的 PyTorch 实现。

## 更新

**2021/03/27:**

(1) 为语义分割发布预训练模型，其中 PointNet++ 可以获得 **53.5\%** mIoU.

(2) 发布分类和部件分割的预训练模型在 `log/` 目录中

**2021/03/20:** 更新用于分类的代码

(1) 增加代码用于训练 **ModelNet10** 数据集，使用配置 ``--num_category 10``.

(2) 增强代码用于在 CPU 上执行，使用配置 ``--use_cpu``.

(3) 增加代码用于离线数据处理，加快训练速度，使用配置 ``--process_data``.

(4) 增加代码用于训练均匀采样，使用配置 ``--use_uniform_sample``.

**2019/11/26:**

(1) 修改先前代码中的部分错误，增加数据增强的技巧。现在分类仅通过1024个点就可以得到 **92.8\%**!

(2) 增加测试代码，包括分类和分割，以及语义分割的可视化。

(3) 组织所有的模型到 `./models` 目录中方便使用。

## 安装

最新代码测试在 Ubuntu 16.04, CUDA10.1, PyTorch 1.6 and Python 3.7:

```shell
conda install pytorch==1.6.0 cudatoolkit=10.1 -c pytorch
```

## 分类 (ModelNet10/40)

### 数据准备（分类）

下载对齐数据 **ModelNet** [here](https://shapenet.cs.stanford.edu/media/modelnet40_normal_resampled.zip)
并且保存在 `data/modelnet40_normal_resampled/`.

### 运行（分类）

下面的代码可以运行在不同的模式下：

* 如果需要离线处理数据，可以在第一次运行时使用 `--process_data` 。也可以下载预处理数据 [here](https://drive.google.com/drive/folders/1_fBYbDO3XSdRt3DSbEBe41r5l9YpIGWF?usp=sharing) 并且保存在 `data/modelnet40_normal_resampled/`.
* 如果需要训练 ModelNet10，可以使用 `--num_category 10`.

```shell
# ModelNet40
## Select different models in ./models 

## e.g., pointnet2_ssg 没有法向特征
python train_classification.py --model pointnet2_cls_ssg --log_dir pointnet2_cls_ssg
python test_classification.py --log_dir pointnet2_cls_ssg

## e.g., pointnet2_ssg 增加法向特征
python train_classification.py --model pointnet2_cls_ssg --use_normals --log_dir pointnet2_cls_ssg_normal
python test_classification.py --use_normals --log_dir pointnet2_cls_ssg_normal

## e.g., pointnet2_ssg 基于均匀采样
python train_classification.py --model pointnet2_cls_ssg --use_uniform_sample --log_dir pointnet2_cls_ssg_fps
python test_classification.py --use_uniform_sample --log_dir pointnet2_cls_ssg_fps

# ModelNet10
## 与 ModelNet40 的配置相同，仅使用 --num_category 10

## e.g., pointnet2_ssg 没有法向特征
python train_classification.py --model pointnet2_cls_ssg --log_dir pointnet2_cls_ssg --num_category 10
python test_classification.py --log_dir pointnet2_cls_ssg --num_category 10
```

### 性能（分类）

| Model | Accuracy |
|--|--|
| PointNet (Official) |  89.2|
| PointNet2 (Official) | 91.9 |
| PointNet (Pytorch without normal) |  90.6|
| PointNet (Pytorch with normal) |  91.4|
| PointNet2_SSG (Pytorch without normal) |  92.2|
| PointNet2_SSG (Pytorch with normal) |  92.4|
| PointNet2_MSG (Pytorch with normal) |  **92.8**|

## 部件分割 (ShapeNet)

### 数据准备（部件分割）

下载对齐的数据 **ShapeNet** [here](https://shapenet.cs.stanford.edu/media/shapenetcore_partanno_segmentation_benchmark_v0_normal.zip)  
保存在 `data/shapenetcore_partanno_segmentation_benchmark_v0_normal/`.

### 执行（部件分割）

```shell
## Check model in ./models 
## e.g., pointnet2_msg
python train_partseg.py --model pointnet2_part_seg_msg --normal --log_dir pointnet2_part_seg_msg
python test_partseg.py --normal --log_dir pointnet2_part_seg_msg
```

### 性能（部件分割）

| Model | Inctance avg IoU| Class avg IoU|
|--|--|--|
|PointNet (Official)|83.7|80.4|
|PointNet2 (Official)|85.1|81.9|
|PointNet (Pytorch)|84.3|81.1|
|PointNet2_SSG (Pytorch)|84.9|81.8|
|PointNet2_MSG (Pytorch)|**85.4**|**82.5**|

## 语义分割 (S3DIS)

### 数据准备（语义分割）

下载三维室内解析数据集 ( 3D indoor parsing dataset，**S3DIS**) [here](http://buildingparser.stanford.edu/dataset.html)  
保存在 `data/s3dis/Stanford3dDataset_v1.2_Aligned_Version/`.

```shell
cd data_utils
python collect_indoor3d_data.py
```

处理数据保存在 `data/stanford_indoor3d/`.

### 执行（语义分割）

```shell
## Check model in ./models 
## e.g., pointnet2_ssg
python train_semseg.py --model pointnet2_sem_seg --test_area 5 --log_dir pointnet2_sem_seg
python test_semseg.py --log_dir pointnet2_sem_seg --test_area 5 --visual
```

可视化结果保存在 `log/sem_seg/pointnet2_sem_seg/visual/`
并且可以使用 [MeshLab](http://www.meshlab.net/) 可视化 .obj 文件

### 性能（语义分割）

|Model  | Overall Acc |Class avg IoU | Checkpoint|
|--|--|--|--|
| PointNet (Pytorch) | 78.9 | 43.7| [40.7MB](log/sem_seg/pointnet_sem_seg) |
| PointNet2_ssg (Pytorch) | **83.0** | **53.5**| [11.2MB](log/sem_seg/pointnet2_sem_seg) |

## 可视化

### 使用 show3d_balls.py

```shell
## build C++ code for visualization
cd visualizer
bash build.sh 
## run one example 
python show3d_balls.py
```

![可视化结果](/visualizer/pic.png)

### 使用 MeshLab

![可视化结果](/visualizer/pic2.png)

## 参考文献

1. [halimacc/pointnet3](https://github.com/halimacc/pointnet3)
1. [fxia22/pointnet.pytorch](https://github.com/fxia22/pointnet.pytorch)
1. [charlesq34/PointNet](https://github.com/charlesq34/pointnet)
1. [charlesq34/PointNet++](https://github.com/charlesq34/pointnet2)

## 引用

如果这个仓库在你的研究中有用，请考虑引用：

```bib
@article{Pytorch_Pointnet_Pointnet2,
      Author = {Xu Yan},
      Title = {Pointnet/Pointnet++ Pytorch},
      Journal = {https://github.com/yanx27/Pointnet_Pointnet2_pytorch},
      Year = {2019}
}
```

```bib
@InProceedings{yan2020pointasnl,
  title={PointASNL: Robust Point Clouds Processing using Nonlocal Neural Networks with Adaptive Sampling},
  author={Yan, Xu and Zheng, Chaoda and Li, Zhen and Wang, Sheng and Cui, Shuguang},
  journal={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  year={2020}
}
```

```bib
@InProceedings{yan2021sparse,
  title={Sparse Single Sweep LiDAR Point Cloud Segmentation via Learning Contextual Shape Priors from Scene Completion},
  author={Yan, Xu and Gao, Jiantao and Li, Jie and Zhang, Ruimao, and Li, Zhen and Huang, Rui and Cui, Shuguang},
  journal={AAAI Conference on Artificial Intelligence ({AAAI})},
  year={2021}
}
```

```bib
@InProceedings{yan20222dpass,
      title={2DPASS: 2D Priors Assisted Semantic Segmentation on LiDAR Point Clouds}, 
      author={Xu Yan and Jiantao Gao and Chaoda Zheng and Chao Zheng and Ruimao Zhang and Shuguang Cui and Zhen Li},
      year={2022},
      journal={ECCV}
}
```
## 使用这个代码库选择项目

* [PointConv: Deep Convolutional Networks on 3D Point Clouds, CVPR'19](https://github.com/Young98CN/pointconv_pytorch)
* [On Isometry Robustness of Deep 3D Point Cloud Models under Adversarial Attacks, CVPR'20](https://github.com/skywalker6174/3d-isometry-robust)
* [Label-Efficient Learning on Point Clouds using Approximate Convex Decompositions, ECCV'20](https://github.com/matheusgadelha/PointCloudLearningACD)
* [PCT: Point Cloud Transformer](https://github.com/MenghaoGuo/PCT)
* [PSNet: Fast Data Structuring for Hierarchical Deep Learning on Point Cloud](https://github.com/lly007/PointStructuringNet)
* [Stratified Transformer for 3D Point Cloud Segmentation, CVPR'22](https://github.com/dvlab-research/stratified-transformer)