This repository contains the code for the paper
CoE: Deep Coupled Embedding for Non-Rigid Point Cloud Correspondences

• Ubuntu 20.04
• Python 3.8

We recommand you install necessary dependencies by conda:

conda create -n coe python=3.8
conda activate coe 
conda install pytorch cudatoolkit -c pytorch 
pip install -r requirements.txt 

Please download the datasets from the following links:

• FAUST
• SCAPE
• SURREAL

and put them in the ./data folder.

The data is organized as follows:

├── data
    ├── FAUST
        ├── off
        ├── corres
    ├── SCAPE
        ├── off
        ├── corres
...

We extend our gratitude to the original dataset providers for their valuable contributions and we acknowledge that full credit belongs to the original authors.

python train.py --config config/scape.json

The pre-computation will be performed automatically and saved in input folder.

The training process will be saved in the saved folder.

python test.py --config config/scape_test.json -d scape -m checkpoints/scape.pth

The evaluation results will be saved in a subfolder under the model path.

And in Matlab, you can run the following code to evaluate the geodesic error results:

eval/eval_scaperm.m

We provide pretrained models for the FAUST, SCAPE and SURREAL dataset. You can find the model in the checkpoints folder.

We provide the visualization code in the visualization folder. You can visualize the correspondence results by running the following command:

python visualization/correspondence.py

The implementation of DiffusionNet is modified from the official implementation.

The cross-attention block is modified from DPFM.

The LBO computation is modified from SSMSM.

We thank the authors for making their codes publicly available.

If you find the code is useful, please cite the following paper

@inproceedings{zeng2024,
    title = {{CoE}: Deep Coupled Embedding for Non-Rigid Point Cloud Correspondences},
    author = {Zeng, Huajian and Gao, Maolin and Cremers, Daniel},
    booktitle = {International Conference on 3D Vision (3DV)},
    month = {March},
    year = {2025},
}