Accompanying code for our IROS 2019 paper
@InProceedings{DTSDF_IROS_2019,
author = {M. {Splietker} and S. {Behnke}},
title = {Directional {TSDF}: Modeling Surface Orientation for Coherent Meshes},
booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
year = {2019},
pages = {1727--1734}
}
This project is deprecated in favor of our new implementation
This code is a fork of the original MeshHashing by Dong et al. Please make sure to cite our and their corresponding papers, if you use the code.
Because the code used GPU accelerated code, it requires OpenCV to be build with WITH_CUDA=ON. You may need to set the OpenCV_DIR variable in CMakeLists.txt.
sudo apt install libglfw3-dev ros-melodic-sophus libglm-dev libcapnp-dev
git submodule update --init --recursive
mkdir build
cd build
cmake ..
make
You might experience some warnings or errors originating from compiling Eigen with CUDA. In that case use a more recent Eigen version (>= 3.3.9).
cd bin
./reconstruction
The program is controlled by the config file config/args.yml and accompanying dataset config specified by the dataset_type field. E.g. dataset_type: 3 corresponds to the config file config/TUM3.yml for datasets in the TUM fr3 format.