PointGMM

This is the official Pytorch implementation of the CVPR2020 paper PointGMM: a Neural GMM Network for Point Clouds.

– Download the ShapeNetCore.v2 dataset.
   Redirect constants.DATASET to the dataset directory.

– Pre-trained models are available here (optional).
   Redirect constants.CHECKPOINTS_ROOT to the models directory.

– Train a VAE model: python train.py -d 0 -c airplane.
   where d specify the GPU id and c specify one of the ShapeNetCore categories.

– Train a registration model: python train.py -d 1 -c chair -r.

– Play with a pre-trained model via eval_ae.py.