GeoSVR: Taming Sparse Voxels for Geometrically Accurate Surface Reconstruction

This repository provides the reconstructed meshes and resources for the paper GeoSVR: Taming Sparse Voxels for Geometrically Accurate Surface Reconstruction, which presents an explicit voxel-based framework for accurate, detailed, and complete surface reconstruction.

• 📚 Paper
• 🌐 Project Page
• 💻 Code

Reconstruction on Tanks and Temples and DTU Datasets

Here we provide the reconstructed meshes of the paper's experiments from GeoSVR.

You can browse all the released meshes at:

• meshes_complete/: The complete meshes of the two datasets.

• DTU_meshes_eval/: The meshes on DTU datasets, with strict filtering strategy for evaluation.

• TnT_meshes_eval/: The meshes on TnT datasets, with strict filtering strategy for evaluation.

Metrics shall be reproduced with the results with postfix of _eval.

Download

from huggingface_hub import snapshot_download
snapshot_download(repo_id="Fictionary/GeoSVR", cache_dir='./GeoSVR/results', local_dir ='./GeoSVR/results')

or use Git to clone this repository with LFS.

Citation

@article{li2025geosvr,
  title={GeoSVR: Taming Sparse Voxels for Geometrically Accurate Surface Reconstruction},
  author={Li, Jiahe and Zhang, Jiawei and Zhang, Youmin and Bai, Xiao and Zheng, Jin and Yu, Xiaohan and Gu, Lin},
  journal={Advances in Neural Information Processing Systems},
  year={2025}
}