PF-Net-Point-Fractal-Network

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://arxiv.org/abs/2003.00410

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 ‘croppointnum’ to control the number of missing points. Change ‘pointscaleslist ’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

4) Visualization of csv File

We provide some incomplete point cloud in file 'testone'. Use the following code to complete a incomplete point cloud of csv file: ``` python Testcsv.py ``` change ‘infile’and ‘infilereal’to select different incomplete point cloud in ‘testone’

5) Visualization of Examples

Using Meshlab to visualize the txt files.