Variational Autoencoding of Dental Point Clouds

Paper | Data

Requirements:

Code was run and tested on

python=3.9
pytorch=1.9.1
h5py=3.10
nflows=0.14
pyntcloud=0.3
scikit-learn=1.3
tqdm=4.66
plyfile=1.0
gitpython=3.1
tensorboard=2.16
setuptools==59.5.0

Metrics are an updated version of SetVAE's metrics for python 3.9. Install them via:

bash ./install.sh

Code was tested on Ubuntu 22.04 using CUDA 11.8.

If anything goes wrong during installation, it can be helpful to delete .cache/torch_extentions folder for a clean build

Data

FDI 16 data can be downloaded here, both as meshes and point clouds.

Checkpoint

Checkpoint can be downloaded from here

Training

VAE training can be run using

python ./main.py --x_train path_to_train_data --x_val path_to_val_data 

Flow prior training can be run using:

python ./main.py --x_train path_to_train_data --x_val path_to_val_data  --x_test path_to_test_data --test_name insert_test_name --seed insert_seed_num

Citation

@misc{ye2024variational,
      title={Variational Autoencoding of Dental Point Clouds}, 
      author={Johan Ziruo Ye and Thomas Ørkild and Peter Lempel Søndergaard and Søren Hauberg},
      year={2024},
      eprint={2307.10895},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}