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.