Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/gkouros/ref-dvgo
Reflection-Aware Direct Voxel Grid Optimization for an Improved Quality-Efficiency Trade-Off in Reflective Scene Reconstruction
https://github.com/gkouros/ref-dvgo
Last synced: about 2 months ago
JSON representation
Reflection-Aware Direct Voxel Grid Optimization for an Improved Quality-Efficiency Trade-Off in Reflective Scene Reconstruction
- Host: GitHub
- URL: https://github.com/gkouros/ref-dvgo
- Owner: gkouros
- License: mit
- Created: 2023-07-28T13:11:05.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2023-09-12T08:06:27.000Z (over 1 year ago)
- Last Synced: 2023-09-12T16:09:53.054Z (over 1 year ago)
- Language: Python
- Size: 281 KB
- Stars: 11
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# ref-dvgo
## 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.
## Installation
Simply run `$ bash install_env.sh` to install the conda environment that is required to run the code.
## Datasets
This codebase is evaluated on two datasets that you can download from the links below:
- [ref-shiny](https://storage.googleapis.com/gresearch/refraw360/ref.zip)
- [smart-car](https://drive.google.com/file/d/16BDhyQJP4mcmFWaJoBBnvyMa1I3lr59o/view?usp=drive_link)
All the scripts assume that the datasets are located in a directory called *data/* at the root of the repository.
## Training
To train a model on a specific scene e.g. the car scene from the blender shiny dataset use the following command:
```shell
$ bash train.py ref_shiny/car
```
You can provide command line arguments that overwrite the config file arguments like below:
```shell
$ 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.
## Acknowledgements
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](https://github.com/sunset1995/DirectVoxGO), [multinerf](https://github.com/google-research/multinerf), and [refnerf-pytorch](https://github.com/gkouros/refnerf-pytorch).
## Citation
```latex
@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}
}
```
```latex
@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}
}
```