KT-Net: Knowledge Transfer for Unpaired 3D Shape Completion

• Preprint paper.

Requirements

• Ubuntu 14.04 or higher
• CUDA 10.0 or higher
• Python v3.7 or higher
• Pytorch v1.2 or higher

Specifically, The code has been tested with:

• Ubuntu 18.04, CUDA 10.2, python 3.8.15, Pytorch 1.6.0, GeForce RTX 2080Ti.

Installation

• Create the conda environment.

  conda create -n kt-net python=3.8
  conda activate kt-net
  

• Intall some packages.

  pip install -r requirements.txt
  

• Install EMD.

  cd net/util/emd_module
  python setup.py install
  cd ../../..
  

Dataset & Pretrained model

• 3DEPN;
• CRN Refer to ShapeInversion;
• Real-World Data Refer to Pcl2Pcl;
• Pretrained Weights. Please place the data to ./dataset and the pretrained model to ./pretrain.

Train && Test

To train the model, you can edit the parameter in the file train_KT.sh and run the command:

sh train_KT.sh

To test the model, you can edit the parameter in the file test_KT.sh and run the command:

sh test_KT.sh

Acknowledgement

The code is in part built on MSC. The original code of emd is rendered from MSN. The original code of chamfer3D is rendered from "chamferDistancePytorch".