This repository contains the source code for 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 use the PUGAN dataset in our experiments, which are available below:

• PUGAN: training h5 file
• PUGAN: testing meshes

To generation the testing point clouds (.xyz files), please refer to the PUGAN repo.

To use our code, make sure that the environment and PyTorch extensions are installed according to the instructions in the main page. Then modify the dataset path in the configuration files.

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

python train.py

To evaluate a pre-trained model, first specify the model_path in configuration file, then run:

python test.py

This repo is based on:

• PUGAN

We thank the authors for their great job!