S-VolSDF

This is the official implementation of S-VolSDF: Sparse Multi-View Stereo Regularization of Neural Implicit Surfaces.

Haoyu Wu, Alexandros Graikos, Dimitris Samaras

International Conference on Computer Vision (ICCV), 2023

Project page | Paper

---

Installation

The code is compatible with python 3.6, CUDA 11.1, and pytorch 1.9.0.

conda create --name s_volsdf python=3.6
conda activate s_volsdf
pip install -r requirements.txt

---

Data

Download data_s_volsdf, inlcuding the DTU and the BlendedMVS Dataset. Then,

unzip data_s_volsdf.zip -d ./

---

Train

# DTU
python runner.py testlist=scan106 outdir=exps_mvs exps_folder=exps_vsdf opt_stepNs=[100000,0,0]
# BlendedMVS (BMVS)
python runner.py vol=bmvs testlist=scan4 outdir=exps_mvs exps_folder=exps_vsdf opt_stepNs=[100000,0,0]

• Train all scans sequentially: testlist='config/lists/dtu.txt' or testlist='config/lists/bmvs.txt'
• Check more details in ./config

---

Eval - 3D Reconstruction

# DTU  
python evals/eval_dtu.py --datadir exps_mvs --scan 106 --data_dir_root data_s_volsdf
# BlendedMVS (BMVS) 
python evals/eval_bmvs.py --datadir exps_mvs --scan 4 --data_dir_root data_s_volsdf

• Evaluate all scans: --scan -1
• The metric is the Chamfer distance (in mm) of the output point clouds .

Eval - Novel View Synthesis

1. Render and save RGBD

# DTU
python eval_vsdf.py --conf dtu --eval_rendering --data_dir_root data_s_volsdf --expname ours --exps_folder exps_vsdf --evals_folder exps_result 
# BMVS: --conf bmvs

2. Simple image-based rendering

# prepare data
python runner.py +create_scene=true outdir=exps_ibr testlist='config/lists/dtu.txt' # DTU
python runner.py vol=bmvs +create_scene=true outdir=exps_ibr testlist='config/lists/bmvs.txt' # BMVS

# image-based rendering
python simple_ibr.py outdir=exps_ibr +evals_folder=exps_result testlist='config/lists/dtu.txt' # DTU
python simple_ibr.py vol=bmvs outdir=exps_ibr +evals_folder=exps_result testlist='config/lists/bmvs.txt' # BMVS

3. Evaluate

# DTU
python eval_vsdf.py --conf dtu --data_dir_root data_s_volsdf --eval_rendering --expname ours --exps_folder exps_vsdf --evals_folder exps_result --result_from blend
# BMVS: --conf bmvs

---

Acknowledgements

• This work was supported in part by the NASA Biodiversity Program (Award 80NSSC21K1027), and NSF Grant IIS-2212046.
• We also thank Alfredo Rivero for his thoughtful feedback and meticulous proofreading.
• We borrowed code from VolSDF, CasMVSNet, UCSNet, TransMVSNet, and DTUeval-python. We thank all the authors for their great work and repos.

---

Citation

If you find our code useful for your research, please cite

@article{wu2023s,
  title={S-VolSDF: Sparse Multi-View Stereo Regularization of Neural Implicit Surfaces},
  author={Wu, Haoyu and Graikos, Alexandros and Samaras, Dimitris},
  journal={arXiv preprint arXiv:2303.17712},
  year={2023}
}