PU-Net: Point Cloud Upsampling Network.
PU-Net is a point cloud upsampling algorithm based on deep learning, which is collected into our open source algorithm library of point cloud. The original project is run on python2 and tensorflow-1.3, which is in folder './src'. In order to make algorithm support multiple frameworks, we attempt to expande the PyTorch version which is in folder './punet_pytorch'. Besides we upgrade original codes with python3, tensorflow-1.14 and CUDA-10, which is in folder './src3'.
Algorithm analysis
please refer to "PU-Net algorithm analysis.doc"
Running project
please refer to the file 'requirements.txt' and README.md under each version.
Performance comparison
We compare the algorithm performance of different frameworks under tensorflow and pytorch respectively. The main evaluation metric is NUC (refer to the paper for details). Note: we take "D=40" in our calculation.
Paper:
p |
0.002 |
0.004 |
0.006 |
0.008 |
0.010 |
0.012 |
NUC |
0.174 |
0.138 |
0.122 |
0.115 |
0.112 |
0.110 |
TF:
p |
0.002 |
0.004 |
0.006 |
0.008 |
0.010 |
0.012 |
0.015 |
NUC |
0.20004781 |
0.16366877 |
0.14697995 |
0.13717446 |
0.13099714 |
0.12588924 |
0.12350797 |
PyTorch:
p |
0.002 |
0.004 |
0.006 |
0.008 |
0.010 |
0.012 |
0.015 |
NUC |
0.17464873 |
0.14155353 |
0.12641019 |
0.11760626 |
0.11187568 |
0.11156724 |
0.11221573 |
Contributors
name: Zhang Yongchi
email: zhangych02@pcl.ac.cn