NeuSurf

Implementation of AAAI'24 paper NeuSurf: On-Surface Priors for Neural Surface Reconstruction from Sparse Input Views

Project Page | Paper | Data | Mesh Results

Overview

Installation

Our code is implemented in Python 3.10, PyTorch 2.0.0 and CUDA 11.7.

• Install Python dependencies

conda create -n neusurf python=3.10
conda activate neusurf
pip install torch==2.0.0 torchvision==0.15.1
pip install -r requirements.txt

• Compile C++ extensions

cd extensions/chamfer_dist
python setup.py install

Dataset

Data structure:

data
|-- DTU_pixelnerf
    |-- <case_name, e.g. dtu_scan24>
        |-- cameras_sphere.npz
        |-- pcd
            |-- <case_name>.ply
        |-- cam4feat
            |-- pair.txt
            |-- cam_00000000_flow3.txt
            |-- cam_00000001_flow3.txt
            ...
        |-- image
            |-- 000000.png
            |-- 000001.png
            ...
        |-- mask
            |-- 000.png
            |-- 001.png
            ...
|-- DTU_sparseneus
|-- blendedmvs_sparse

You can directly download the processed data here.

Running

• Training

CUDA_VISIBLE_DEVICES=0
python exp_runner.py --mode train --conf ./confs/dtu.conf --case <case_name, e.g. dtu_scan24>

• Extract mesh

CUDA_VISIBLE_DEVICES=0
python exp_runner.py --mode validate_mesh --conf ./confs/dtu.conf --case <case_name> --is_continue

Citation

If you find our work useful in your research, please consider citing:

@inproceedings{huang2024neusurf,
    title={NeuSurf: On-Surface Priors for Neural Surface Reconstruction from Sparse Input Views},
    author={Huang, Han and Wu, Yulun and Zhou, Junsheng and Gao, Ge and Gu, Ming and Liu, Yu-Shen},
    booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
    volume={38},
    number={3},
    pages={2312--2320},
    year={2024}
}

Acknowledgement

This implementation is based on CAP-UDF, D-NeuS and Vis-MVSNet. Thanks for these great works.