Ref-DVGO: Reflection-Aware Direct Voxel Grid Optimization for an Improved Quality-Efficiency Trade-Off in Reflective Scene Reconstruction
Source code of the paper titled "Reflection-Aware Direct Voxel Grid Optimization for an Improved Quality-Efficiency Trade-Off in Reflective Scene Reconstruction", to be presented at the TRICKY workshop of ICCV 2023.
Simply run $ bash install_env.sh
to install the conda environment that is required to run the code.
This codebase is evaluated on two datasets that you can download from the links below:
All the scripts assume that the datasets are located in a directory called data/ at the root of the repository.
To train a model on a specific scene e.g. the car scene from the blender shiny dataset use the following command:
$ bash train.py ref_shiny/car <JOBID> <Optional ARGS>
You can provide command line arguments that overwrite the config file arguments like below:
$ bash train.py ref_shiny/car 12345 fine_train.N_rand=4096 fine_train.N_iters=80000
Training with the default configuration takes around 30 minutes on average on an Nvidia TITAN V.
We gratefully acknowledge funding support from the Sim2Real2 project, in the context of the Ford-KU Leuven alliance program.
This codebase is heavily based on dvgo, multinerf, and refnerf-pytorch.
@misc{kouros2023refdvgo,
title={Ref-DVGO: Reflection-Aware Direct Voxel Grid Optimization for an Improved Quality-Efficiency Trade-Off in Reflective Scene Reconstruction},
author={Georgios Kouros and Minye Wu and Shubham Shrivastava and Sushruth Nagesh and Punarjay Chakravarty and Tinne Tuytelaars},
year={2023},
eprint={2308.08530},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2308.08530}
}
@misc{kouros2023refdvgo,
author={Georgios Kouros and Minye Wu and Shubham Shrivastava and Sushruth Nagesh and Punarjay Chakravarty and Tinne Tuytelaars},
title={Reflection-Aware Direct Voxel Grid Optimization for an Improved Quality-Efficiency Trade-Off in Reflective Scene Reconstruction},
year={2023},
url={https://github.com/gkouros/ref-dvgo}
}