• Semantic Point Cloud Upsampling (SPU)
• • Environment
  • Pretrained Model
  • Trainning
  • Test
  • Performance
  • • PointNet:
    • DGCNN:
  • Cite
  • contributors

Semantic Point Cloud Upsampling (SPU)

This is the code project of Semantic Point Cloud Upsampling (SPU) published on IEEE Transactions on Multimedia.

Environment

Python 3.6

PyTorch 1.2.0

Torchvision 0.4.0

KNN_CUDA 0.2.0

Pretrained Model

You can get the trained upsampling model in the dir of ./savedModel.

The pre-trained classification networks are at ./PretrainModel.

Trainning

You can switch DGCNN or PointNet as classification network.

cd ./train
python trainSPUDGCNN8x.py
python trainSPUPointNet8x.py

Test

In this commond, you can revise upsampling factor to achieve paper results.

cd ./test
python TestSPU2.py

Performance

PointNet:

scale	CD	HD	Top1 acc	Cls ave	F-Euc
4X	0.002771	0.021539	0.883712	0.842134	3.395767
8X	0.003900	0.038486	0.845219	0.803791	5.123443

DGCNN:

scale	CD	HD	Top1 acc	Cls ave	F-Euc
4X	0.002689	0.023840	0.868314	0.832221	12.505258
8X	0.003910	0.042285	0.761345	0.725174	14.432164

Cite

Zhuangzi Li, Ge Li, Thomas Li, Shan Liu, Wei Gao, “Semantic Point Cloud Upsampling,” IEEE Transactions on Multimedia (TMM), accepted in March 2022.

contributors

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