This module generates the TSDF volumes based on the house.obj file and some camera poses.
First step is to download and install HDF5, which we use to store the resulting data in a compressed format.
First you have to download this file:
https://support.hdfgroup.org/ftp/HDF5/releases/hdf5-1.10/hdf5-1.10.6/src/CMake-hdf5-1.10.6.tar.gz
Unzip and execute the CMake-hdf5-1.10.6/build-unix.sh file, this will take a few minutes.
After running through without any errors, there will be a file with the name: HDF5-1.10.6-Linux.tar.gz.
This file contains the complete build, including the include, lib, bin and share folder.
Unpack this folder to a new location.
In it you will find this folder: HDF5-1.10.6-Linux/HDF_Group/HDF5/1.10.6/share/cmake/hdf5, which has to be set in the CMakeLists.txt file as the HDF5_DIR.
In the second step you need to install TCLAP.
Download the source files, we used version: 1.2.2.
Unzip the downloaded file. And update the path in the CMakeLists.txt.
After downloading and updating the paths, you only have to build the current project with the given CMakeLists.txt.
mkdir cmake
cd cmake
cmake -DCMAKE_BUILD_TYPE=RELEASE ..
make -j 8You need a few things to start a TSDF generation run.
You need an object file generated (for example generated via the SUNCG folder) and also a cameraposition file also generated via the SUNCGToolBox After that:6
./sdfgen -o {SUNCG_FOLDER}/house/10704e82d0ef2bf37a658af4fb81c06c/house.obj -c {SUNCG_FOLDER}/house/10704e82d0ef2bf37a658af4fb81c06c/camerapositions -r 128 -f output_folder
SDFGen has a lot of tuneable hyperparameters,
Again, we provide here also a script to do this automatically for the generated house.obj and camerapositions.
Please, change the paths in generate_tsdf_volumes.py, so that it will convert all generate house.obj and camera positions into TSDF voxelgrids.
To view these voxelgrids one can use the TSDFRenderer.