UniPCGC

<h1 align="center">UniPCGC: Towards Practical Point Cloud Geometry Compression via an Efficient Unified Approach</h1> <p align="center"> <strong>Kangli Wang</strong>¹, <strong>Wei Gao</strong>^1,2*<br> (<em>* Corresponding author</em>) </p> <p align="center"> ¹SECE, Peking University<br> ²Peng Cheng Laboratory, Shenzhen, China </p> <p align="center"> <a href="https://arxiv.org/abs/2503.18541"><img src="https://img.shields.io/badge/Arxiv-2503.18541-b31b1b.svg?logo=arXiv" alt="arXiv"></a> <a href="https://github.com/Wangkkklll/UniPCGC?tab=MIT-1-ov-file"><img src="https://img.shields.io/badge/License-MIT-yellow" alt="License"></a> <a href="https://uni-pcgc.github.io/"><img src="https://img.shields.io/badge/Project_Page-UniPCGC-blue.svg" alt="Home Page"></a> </p>

📣 News

• [24-12-09] 🔥 Our paper has been accepted to AAAI 2025.
• [25-03-08] 🔥 We release lossless compression code.
• [25-06-29] 🔥 We release all UniPCGC codes and checkpoints.

Todo

• [x] ~~Release training code~~
• [x] ~~Release inference code~~
• [x] ~~Release the Paper~~
• [x] ~~Release checkpoint~~
• [x] ~~Simplify the code~~

Links

Our work on gaussian compression has also been released. Welcome to check it.

• 🔥 GausPcc [Arxiv'25]: Efficient 3D Gaussian Compression ! [Arxiv] [Project]

Our work on any source point cloud compression has also been released. Its performance is better than UniPCGC. Welcome to check it.

• 🔥 AnyPcc [CVPR'26]: Any Source Point Cloud Compression ! [Arxiv] [Project]

📌 Introduction

We propose an efficient unified point cloud geometry compression framework UniPCGC. It is a lightweight framework that supports lossy compression, lossless compression, variable rate and variable complexity. First, we introduce the Uneven 8-Stage Lossless Coder (UELC) in the lossless mode, which allocates more computational complexity to groups with higher coding difficulty, and merges groups with lower coding difficulty. Second, Variable Rate and Complexity Module (VRCM) is achieved in the lossy mode through joint adoption of a rate modulation module and dynamic sparse convolution. Finally, through the dynamic combination of UELC and VRCM, we achieve lossy compression, lossless compression, variable rate and complexity within a unified framework. Compared to the previous state-of-the-art method, our method achieves a compression ratio (CR) gain of 8.1% on lossless compression, and a Bjontegaard Delta Rate (BD-Rate) gain of 14.02% on lossy compression, while also supporting variable rate and variable complexity.

<div align="center"> <img src="assets/unipcgc.jpg" width = 75% height = 75%/> <br> Ilustration of the proposed UniPCGC framework. </div>

🔑 Setup

Our cuda version is 11.6. We recommend using the following command to complete the installation.

conda create -n unipcgc python=3.8
conda activate unipcgc
pip install -r requirements

After completing the above installation, you should also install MinkowskiEngine. For the installation of MinkowskiEngine, see the official repository.

Warn!!! The following yml environment is in the development version environment, which introduces many irrelevant libraries and is not recommended. Type the command for general installation

conda env create -f environment.yml

🧩 Dataset Preparation and Pretrained Model

Please refer to the following links to obtain the data. We thank these great works. | Datasets | Download Link | |:-----: |:-----: | | ShapeNet | Link | | 8iVFB | Link | | Testdata | Baidu Netdisk (kkll) |

Please refer to the following links to obtain the pretrained models. | Models | Download Link | |:-----: |:-----: | | UniPCGC | Link |

🚀 Running

For lossless compression, run the following code to train

python train_lossless.py --dataset "your dataset dir" --lr 8e-4

run the following code to compress and decompress

python unicoder_lossless.py --filedir "your dataset dir" --ckptdir "your ckpt dir"

For more training scripts, please refer to scripts.

🔎 Contact

If your have any comments or questions, feel free to contact kangliwang@stu.pku.edu.cn.

👍 Acknowledgement

Thanks for their awesome works (PCGCv2 and MinkowskiEngine).

📘 Citation

Please consider citing our work as follows if it is helpful.

@article{wang2025unipcgc,
title={UniPCGC: Towards Practical Point Cloud Geometry Compression via an Efficient Unified Approach},
volume={39},
url={https://ojs.aaai.org/index.php/AAAI/article/view/33387},
DOI={10.1609/aaai.v39i12.33387},
number={12},
journal={Proceedings of the AAAI Conference on Artificial Intelligence},
author={Wang, Kangli and Gao, Wei},
year={2025},
month={Apr.},
pages={12721-12729} }