This repository contains the source code of point cloud auto-encoding and generation of the papers:

1. Snowflake Point Deconvolution for Point Cloud Completion and Generation with Skip-Transformer (TPAMI 2023)

2. SnowflakeNet: Point Cloud Completion by Snowflake Point Deconvolution with Skip-Transformer (ICCV 2021, Oral)

We follow diffusion-point-cloud and evaluate the point cloud generation quality of SPD on the ShapeNet dataset.

• ShapeNet

To use our code, make sure that the environment and PyTorch extensions are installed according to the instructions in the main page.

To train a point cloud generation model from scratch, run:

export CUDA_VISIBLE_DEVICES='0'

python train_ae.py # point cloud auto-encoding
# or 
python train_gen.py # point cloud generation

To evaluate a pre-trained model, first specify the ckpt path, then run:

export CUDA_VISIBLE_DEVICES='0'

python test_ae.py # point cloud auto-encoding
# or 
python test_gen.py # point cloud generation

This repo is based on:

• diffusion-point-cloud

We thank the authors for their great job!