Diffusing Winding Gradients (DWG): A Parallel and Scalable Method for 3D Reconstruction from Unoriented Point Clouds

Building

Please install CMake

Dependencies

• CGAL-5.6.1
• PCL
• Eigen
• CUDA (We tested on CUDA 11.1 and 11.3)

Conda

You can install dependencies with conda (and even compile):

conda create -n DWG
conda activate DWG
conda install conda-forge::mamba
## Maybe skip the last three packages: `conda-forge::cuda conda-forge::llvm-openmp conda-forge::clangxx`
## The system cuda should be used to match the installed GPU driver.
## On Windows, try the default compiler rather than llvm-openmp and clangxx.
mamba install conda-forge::cgal "conda-forge::pcl>1.11" conda-forge::eigen conda-forge::gmp conda-forge::cmake conda-forge::cuda conda-forge::llvm-openmp conda-forge::clangxx
cmake -B build-clang -D CMAKE_BUILD_TYPE=Release -D CMAKE_POLICY_VERSION_MINIMUM=3.5 -D CMAKE_C_COMPILER=clang -D CMAKE_CXX_COMPILER=clang++
## If using the system compiler: cmake -B build-clang -D CMAKE_BUILD_TYPE=Release -D CMAKE_POLICY_VERSION_MINIMUM=3.5
cmake --build build-clang

Windows 10

mkdir build && cd build
cmake .. -G"Visual Studio 16 2019"
cmake --build . --config Release

Linux

mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make

Thanks to libigl, Geometry Central, cudaKDTree, marching-cubes-with-CUDA, these libraries provide us with big assistance.

Run

./DWG_CUDA --in_path <path> --in_name <file name> --out_path <path>

See --help for the function of each flag ./DWG_CUDA --help

TODO

• Provide a more detailed explanation of the parameters.
• The results of small-scale models on Linux are a bit rougher compared to those on Windows. We will identify the problems as soon as possible.
• Clean the code.