This repository includes the source code of the paper DynoSurf: Neural Deformation-based Temporally Consistent Dynamic Surface Reconstruction (ECCV 2024).
Authors: Yuxin Yao, Siyu Ren, Junhui Hou, Zhi Deng, Juyong Zhang, Wenping Wang.
git clone https://github.com/yaoyx689/DynoSurf.git
cd DynoSurf
conda create -n dynosurf python=3.9
conda activate dynosurf
# install pytorch (https://pytorch.org/)
pip install torch==1.13.0+cu116 torchvision==0.14.0+cu116 torchaudio==0.13.0 --extra-index-url https://download.pytorch.org/whl/cu116
# install pytorch3d (https://github.com/facebookresearch/pytorch3d/blob/main/INSTALL.md)
pip install "git+https://github.com/facebookresearch/pytorch3d.git"
# install kaolin (https://kaolin.readthedocs.io/en/latest/notes/installation.html)
pip install kaolin==0.14.0 -f https://nvidia-kaolin.s3.us-east-2.amazonaws.com/torch-1.13.0_cu116.html
# install other dependencies
pip install -r requirements.txt
cd scripts
python run.py
If the input point cloud sequence does not provide normals, put the folder of them under data_source/raw_data/
, compute the normals for them and write them into data_source/[foldername]
. If the normals are already provided, skip this step.
cd process_data
python process_raw_data.py [folder_name]
Write the input data into a format that can be read by the code.
python gene_json.py [folder_name]
Execute training. Change the folder_name
in scripts/run.sh
and run
cd scripts
python run.py
Resolution: If you want to adjust the resolution of the reconstructed surface, you can adjust the parameter tet_grid_volume
(around $10^{-8}$ ~ $10^{-6}$) in confs/base.conf
, which controls the resolution of the generated tetrahedron. The smaller the value, the higher the resolution.
Template Frame: If you want to specify the frame index corresponding to the template surface, please modify the template_idx
in confs/base.conf
.
If you find our code or paper helps, please consider citing:
@inproceedings{yao2024dynosurf,
author = {Yao, Yuxin and Ren, Siyu and Hou, Junhui and Deng, Zhi and Zhang, Juyong and Wang, Wenping},
title = {DynoSurf: Neural Deformation-based Temporally Consistent Dynamic Surface Reconstruction},
booktitle = {European Conference on Computer Vision},
year = {2024},
}