Denoising Diffusion Probabilistic Model for Point Cloud Compression at Low Bit-Rates

Pytorch implementation of the paper "Denoising Diffusion Probabilistic Model for Point Cloud Compression at Low Bit-Rates", published at ICME 2025. This repository is based on diffusion-point-cloud.

ArXiv version

Abstract

Efficient compression of low-bit-rate point clouds is critical for bandwidth-constrained applications. However, existing techniques mainly focus on high-fidelity reconstruction, requiring many bits for compression. This paper proposes a “Denoising Diffusion Probabilistic Model” (DDPM) architecture for point cloud compression (DDPM-PCC) at low bit-rates. A PointNet encoder produces the condition vector for the generation, which is then quantized via a learnable vector quantizer. This configuration allows to achieve a low bitrates while preserving quality. Experiments on ShapeNet and ModelNet40 show improved rate-distortion at low rates compared to standardized and state-of-the-art approaches

Usage

Environment

• conda env create -f env.yml
• conda activate point

Eval

Download our pretrained model in the DDPM-PCC directory from here.

• Extract results_ddpm.zip

DDPM-PCC
│   README.md
│  
└───results
│   └───shapenet
│       │   ...
│   └───modelnet
│       │   ...
│   
└───src
    │   train.py
    │   ...

• Run:

sh run_all_eval.sh

Train

Download Shapenet and ModelNet datasets in DDPM-PCC directory from here.

• Extract datasets.zip

DDPM-PCC
│   README.md
│  
└───datasets
│   └───modelnet40_ply_hdf5_2048/
│       │   ...
│   └───shapenet.hdf5
│   
└───src
    │   train.py
    │   ...

• Run: ShapeNet

cd src

python train.py \
--dataset-path ../datasets/data/shapenet.hdf5 \
--dataset shapenet \
--latent-dim 256 \
--num-steps 200 \
--rotate 0 \
--save-dir ../results/shapenet/vq_diffusion/ae_all_pointnet_bs128 \
--encoder pointnet  \
--train-batch-size 128 \
--val-batch-size 1 \
--num-codecs 128 \
--dim-codecs 32 \
--vq-alpha 2.5

Results

• Quantitavie Results

• Qualitative Results