CP-Net

Cosmos Propagation Network: Deep learning model for point cloud completion

This repository is still under constructions.

If you have any questions about the code, please contact me. Thanks!

Usage

1) Recommended Envrionment & prerequisites

• Pytorch 1.2.0 (1.0.1 and 1.5.0 also works)
• CUDA 10.0
• Python 3.7
• Visdom (optional)
• Open3D(optional for load different dataset)

2) Compile

Compile extension modules from msn for Evaluate the Performance with EMD and F1 socre (No need if only use Chamfer Distance):

git clone https://github.com/Colin97/MSN-Point-Cloud-Completion
cd emd
python3 setup.py install
cd expansion_penalty
python3 setup.py install
cd MDS
python3 setup.py install

And add the related path to show_CD.py

ShapenetPart dataset

  cd dataset
  bash download_shapenet_part16_catagories.sh
  You can also download the dataset from 
  链接：https://pan.baidu.com/s/1MavAO_GHa0a6BZh4Oaogug 提取码：3hoe 

ShapeNet, Compeletion3D are available below:

• ShapeNet
• Completion3D

Train

python Train_FPNet.py 

Change ‘crop_point_num’ to control the number of missing points. Change ‘D_choose’to control without using D-net.

Evaluate the Performance on ShapeNet and other datasets

python show_recon.py

Show the completion results, the program will generate txt files in 'test-examples'.

python show_CD.py

Show the Chamfer Distance, EMD and F1.

Visualization of csv File

We provide some incomplete point cloud in file 'test_one'. Use the following code to complete a incomplete point cloud of csv file:

python Test_csv.py

change ‘infile’and ‘infile_real’to select different incomplete point cloud in ‘test_one’

Visualization of Examples

Using Meshlab or Cloudcompare to visualize the txt/csv files.

Acknowledgemets

This project is built upon [PF-Net-Point-Fractal-Network], [MSN-Point-Cloud-Completion] and [DGCNN]

Others

The program is not beautiful. If you have any questions, please feel free to contact me at the address below. [email protected]

License

This project is open sourced under MIT license.