Environment

The code was tested with python 3.8 and pytorch 1.9 on linux.

If you use a system pip installation, run pip install -r requirements.txt

Then build pytorch extensions

# PointNet++
pip install "git+git://github.com/erikwijmans/Pointnet2_PyTorch.git#egg=pointnet2_ops&subdirectory=pointnet2_ops_lib"
# GPU kNN
pip install --upgrade https://github.com/unlimblue/KNN_CUDA/releases/download/0.2/KNN_CUDA-0.2-py3-none-any.whl
# Customized Extensions
python setup.py build_ext --inplace

Datasets

We use ShapeNet as our training and testing dataset.

To build the dataset, you need to follow OccNet repository to fetch watertight meshes first.

And then, run bash scripts/data_preprocessing/data_prepare.sh. Inside the script file, you need to fill in your own data path.

Training

There are config files in configs folder, with which you can customize your own settings. An important note is that you need to set _CN.dataset.sample_data_root in the config file as your own dataset path.

After setting up the config file, you can start training. Here are some sample scripts, note that inside the train_mis_chair_128-32-16.sh, you will need to set vis_ckpt_path as the path to the trained checkpoint file of the visible training phase.:

For shape completion:

# Training for visible regions
bash scripts/completion/train_vis_chair_128-32-16.sh
# Training for missing regions
bash scripts/completion/train_mis_chair_128-32-16.sh

For shape reconstruction:

Prediction

We offer basic scripts for running prediction. Within the scripts, you need to set ckpt_path and pred_list to provide the path to the trained checkpoint and the path to the file of predicting list.

Run bash scripts/completion/pred_vis_chair_128-64-32.sh for encoding phase predicting

Run bash scripts/completion/pred_mis_chair_128-64-32.sh for decoding phase predicting