Score-PA: Score-based 3D Part Assembly

This is the official implementation of the paper "Score-PA: Score-based 3D Part Assembly", BMVC 2023 (Oral).

Authors: Junfeng Cheng, Mingdong Wu, Ruiyuan Zhang, Guanqi Zhan, Chao Wu and Hao Dong.

• Paper Link

Installation

Requirements

• Python 3.8.5
• PyTorch 1.9.1 + CUDA 11.1
• PyTorch Geomtric 2.0.3
• PyTorch3D 0.6.1

Install Chamfer Distance library

cd ./cd
python setup.py install

Datasets

We follow Generative 3D Part Assembly via Dynamic Graph Learning to use their preprocessed datasets. Here is the download link.

Usage

We have provided the training and testing scripts for the three datasets (chair, table and lamp) in the folder "scripts". You can directly run the scripts to train and test the model. Please strictly use the parameters in the scripts to reproduce the results in the paper.

Training

sh train_{chair / table / lamp}.sh

Testing

sh test_{chair / table / lamp}.sh

Pretrained Models

We release our pretrained models in the "pretrained_models" folder.

Citation

@article{cheng2023score,
  title={Score-PA: Score-based 3D Part Assembly},
  author={Cheng, Junfeng and Wu, Mingdong and Zhang, Ruiyuan and Zhan, Guanqi and Wu, Chao and Dong, Hao},
  journal={British Machine Vision Conference (BMVC)},
  year={2023}
}

Acknowledgement

Our code is inspired by Score-based Generative Modeling through Stochastic Differential Equations and Generative 3D Part Assembly via Dynamic Graph Learning. We thank the authors for their great works.