This is the reference implementation for thin structure reconstruction from an image sequence.
This project is mainly developed and tested at Manjaro Linux. Please refer to the following steps to accomplish your installation.
git clone https://github.com/Totoro97/Vid2Curve.git
cd Vid2Curve
mkdir build
cd build
# ----- Default compile setting -----
cmake ..
make Display
# ----- Or compile with GUI(Pangolin) -----
cmake .. -DUSE_GUI=1
make Display
At the build
directory, run command ./Display
.
In the config.ini
file, you can modify the DataPath
field for your own use and test.
The mesh and the curve skeleton will be saved as curves_mesh.obj
and curves.obj
at the build
directory after the reconstruction process is done. And the camera poses will be saved at cameras.txt
as rotation matrix and translation vectors (from world to camera).
In this repository we have prepared two simple examples for your quick validation. For more cases, please refer to the link.
If you would like to use the project for reconstruction of your own thin objects, please prepare the data as the following steps:
500x500
and 1000x1000
.local_config.ini
file that indicates the internal parameters of the camera and some other information (as the examples show).If you would like to use this project for your own use, please cite:
@article{wang2020vid2curve,
title={Vid2Curve: Simultaneous Camera Motion Estimation and Thin Structure Reconstruction from an RGB Video},
author={Wang, Peng and Liu, Lingjie and Chen, Nenglun and Chu, Hung-Kuo and Theobalt, Christian and Wang, Wenping},
Journal={ACM Trans. Graph. (SIGGRAPH)},
year={2020},
Number={4},
Volume={38},
DOI={https://doi.org/10.1145/3386569.3392476},
Publisher={ACM}
}
This project also uses the codes from :
Thanks for these great implementations.