Seeing Through the Glass: Neural 3D Reconstruction of Object Inside a Transparent Container

July 23, 2026 · View on GitHub

Jinguang Tong, Sundaram Muthu, Fahira Afzal Maken, Chuong Nguyen, Hongdong Li

Teaser image

This repository contains the official implementation of Seeing Through the Glass: Neural 3D Reconstruction of Object Inside a Transparent Container (CVPR2023).

Installation

# Clone this repository
git clone https://github.com/hirotong/refractive-neus

# Create and activate the conda environment
conda env create --file environment.yml
conda activate reneus

python setup.py install

Training

Basic Training

To train a scene:

python train.py --conf configs/base.conf --case $(case name)/withbox

Testing

python train.py --mode evaluate_mesh --conf configs/base.conf --case $(case name)/withbox -r 512 --n_points 10000

Additional Command Line Arguments for train.py

-r

Resolution used to run marching cube algorithm.

--n_points Number of points used for evaluation.

Experiments

Download and uncompress our datset.

Running Experiments

Train the synthetic dataset with SIREN.

bash scripts/train_siren.sh

Acknowledgements

This project builds upon the excellent work of NeuS. We extend our gratitude to all the authors for their valuable contributions to the research community.

Citation

If you find our code or paper helpful for your research, please consider citing our work:

@inproceedings{tong2023seeing,
  title={Seeing through the glass: Neural 3d reconstruction of object inside a transparent container},
  author={Tong, Jinguang and Muthu, Sundaram and Maken, Fahira Afzal and Nguyen, Chuong and Li, Hongdong},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={12555--12564},
  year={2023}
}

Licensing and Patent Notice

This code is released under a CSIRO Non-Commercial License (based on BSD 3-Clause-Clear).

This software may be covered by one or more patents held by CSIRO. Use of this software for commercial purposes may require a license. If you intend to use this software in a commercial context, or believe you may otherwise require a patent license, please contact CSIRO to discuss licensing options: EnquiriesTeam@csiro.au