PF-Net-Point-Fractal-Network

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

This repository is still under constructions.

##0) Environment Pytorch 1.0.1 Python 3.7.4

##1) Dataset

  cd dataset
  bash download_shapenet_part16_catagories.sh

##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_FPNet.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 Use Meshlab to visualize the txt files.

tyronedong/PF-Net-Point-Fractal-Network

PF-Net-Point-Fractal-Network