OpenPointCloud/pointnet&pointnet++/PyTorch

• PointNet && PointNet++
• • Environment
  • Dataset
  • Training
  • Test
  • Performance comparison (IOU)
  • Citation
  • contributors

PointNet && PointNet++

This is the repository of PointNet and PointNet++. The original code is implemented by Tensorflow, while we provide PyTorch version of PointNet and PointNet++ in this project.

Note: we only give the illustration of PyTorch version, and please find the original TF code in ./TF.

Environment

Python 3.6

PyTorch 1.2.0

CUDA 10.2

Dataset

You can downlown the data from here, and then unzip it and rename as ./data.

Training

For pointnet,

cd ./pointnet
python  train_segmentation.py --mode train 

For pointnet++,

cd ./pointnet2
python setup.py install
python  train_segmentation2.py --mode train 

Test

For pointnet,

bash test.sh  

For pointnet++,

bash test.sh

You also can choose different categories to evaluate performance under IOU

Performance comparison (IOU)

Paper:

Chair	Bag	Cap	Car	Guitar	Knife	Lamp	Laptop
89.6	78.7	82.5	74.9	91.5	85.9	80.8	95.3

pointnet:

Chair	Bag	Cap	Car	Guitar	Knife	Lamp	Laptop
88.6	70.5	72.9	72.1	90.2	81.9	76.6	94.5

pointnet++:

Chair	Bag	Cap	Car	Guitar	Knife	Lamp	Laptop
89.6	78.2	76.4	76.0	90.0	83.3	80.4	94.9

Citation

Qi, Charles R and Su, Hao and Mo, Kaichun and Guibas, Leonidas J, “PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation,” arXiv preprint arXiv:1612.00593.

Qi, Charles R and Yi, Li and Su, Hao and Guibas, Leonidas J, "Qi, Charles R and Yi, Li and Su, Hao and Guibas, Leonidas J," arXiv preprint arXiv:1706.02413.

contributors

name: Zhang Yongchi
email: zhangych02@pcl.ac.cn