The code in this repository implements a classical mesh refinement method that minimizes the ZNCC photometric error among pairs of images. It takes a json file in the format defined by openMVG here.
It is the backbone of the paper: Romanoni, Andrea, and Matteo Matteucci. "Mesh-based camera pairs selection and occlusion-aware masking for mesh refinement." Pattern Recognition Letters 125 (2019): 364-372. If you use it please cite:
@article{romanoni2019mesh,
title={Mesh-based camera pairs selection and occlusion-aware masking for mesh refinement},
author={Romanoni, Andrea and Matteucci, Matteo},
journal={Pattern Recognition Letters},
volume={125},
pages={364--372},
year={2019},
publisher={Elsevier}
}
The code is implemented in C++ and OpenGL. You can run the following in Ubuntu/Debian to install the required dependencies
sudo apt install libboost-filesystem-dev libboost-thread-dev libboost-system-dev libopencv-dev libglew-dev libcgal-dev freeglut3-dev libmpfr-dev libglu1-mesa-dev
Clone the repository
git clone --recurse-submodules https://github.com/andresax/photometricRefinement
Compile GLFW
cd external/glfw/
mkdir build/
cd build
cmake ..
make
Compile TransforMesh
cd ../../TransforMesh/
mkdir build/
cd build
cmake ..
make
Compile Photometric Refinement:
From the partent folder of the project:
mkdir build/
cd build
cmake ..
make
Run the example.
From the partent folder of the project:
First uncompress the json (with the output of openMVG):
unzip sfm_data.zip sfm_data.json
Then, execute the program with the configuration file as parameter.
./build/photometricRefinement ./config/configDatasetCastle.conf
The Results will be saved into the folder Result->Result_Castle_* where * will be replaced by the values of the parameter of the current configuration.