This repository is still under constructions.

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

This is the Pytorch implement of CVPR2020 PF-Net: Point Fractal Network for 3D Point Cloud Completion.

https://openaccess.thecvf.com/content_CVPR_2020/papers/Huang_PF-Net_Point_Fractal_Network_for_3D_Point_Cloud_Completion_CVPR_2020_paper.pdf

##0) Environment Pytorch 1.0.1 Python 3.7.4

##1) 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 

##2) Train

python Train_FPNet.py 

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

##3) Evaluate the Performance on ShapeNet

python show_recon.py

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

python show_CD.py

Show the Chamfer Distances and two metrics in our paper.

##4) 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’

##5) Visualization of Examples

Using Meshlab to visualize the txt files.