peterzhousz / paraheat Goto Github PK

ParaHeat: Parallel and Scalable Heat Methods for Geodesic Distance Computation

This is the source code for the algorithms in the following paper:

• Jiong Tao, Juyong Zhang, Bailin Deng, Zheng Fang, Yue Peng, Ying He. 2019. Parallel and Scalable Heat Methods for Geodesic Distance Computation. arXiv:1812.06060.

Compling the code

1. The code has been tested on the following systems:

   • Windows 10 with Microsoft Visual Studio Community 2017;
   • macOS Mojave 10.14.3 with Xcode 10.1 and Homebrew GCC 8.1.0;
   • Ubuntu 18.04 with GCC 7.3.0.

2. The code implements the following three commands:

   • GeodDistSolver for computing geodesic distance;
   • ViewScalarField for visualizing the distance on a mesh;
   • CompareDistance for computing mean relative error of the computed distance.

3. The code requires [Eigen] (http://http://eigen.tuxfamily.org).

   • On Ubuntu or Debian, this can be installed using the following command:

       $ apt-get instal libeigen3-dev
     

   • On macOS, Eigen can be installed using Homebrew command:

       $ brew install eigen
     

   • On Windows, you can download Eigen and extract it into the external folder. Make sure you rename the Eigen folder as eigen, so that you can locate the file external/eigen/Eigen/Dense.

4. In addition, compiling ViewScalarField requires libigl, Glad, and glfw:

   • Download libigl from github, unzip it and rename the folder as libigl, and place it inside the external folder. Make sure you can locate the folder external/libigl/include.
   • Download the Glad library (glad.zip) from its homepage, unzip and place it inside the external folder. Make sure you can locate the folders external/glad/include and external/glad/src.
   • To install glfw:
     • On Ubuntu or Debian, use command apt-get install libglfw3-dev.
     • On macOs, use command brew install glfw.
     • On Windows, download the binary from the homepage and install.
   • Warning: the latest libigl library relies on features of glfw 3.3, and does not compile on older versions of glfw. To avoid compilation errors, you can disable ViewScalarField using the switch WITH_VIEWER during cmake configuration.

5. After installing all the dependencies, use cmake to generate the project file and compile. Create a folder build within the root folder of source code, then run

   $ cmake -DCMAKE_BUILD_TYPE=Release ..
   $ make
   

   If you use a cmake user interface (using the ccmake command for the curses interface, or using a GUI), you can turn on/off OpenMP support and the compilation of the viewer before compling the code.

   • The default compiler on macOS (Apple Clang from Xcode) does not support OpenMP. To achieve best performance, it is recommended to compile the code with Homebrew GCC. For example, if you create a build folder in the source code directory, then use the following command (assuming gcc-8 and g++-8 are installed by Homebrew)

       $ cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=gcc-8 -DCMAKE_CXX_COMPILER=g++-8 ..
     

Usage of commands

1. To compute geodesic distance, use the command

    $ GeodDistSolver PARAMETERS_FILE MESH_FILE DISTANCE_FILE
   

   • PARAMETERS_FILE: the solver parameter file; an example file (SolverParams.txt) is provided.
   • MESH_FILE: the triangle mesh file.
   • DISTANCE_FILE: an output file that stores the distance for each vertex.

   The command will print out peak memory consumption at the end.

2. To visualize the geodesic distance, use the command

    $ ViewScalarField MESH_FILE DATA_FILE
   

   • MESH_FILE: the triangle mesh file.
   • DATA_FILE: a distance file generated by the command GeodDistSolver.

3. To compute distance error

    $ CompareDistance PARAMETERS_FILE DISTANCE_FILE REFERENCE_DISTANCE_FILE
   

   • PARAMETERS_FILE: the parameter file previously used for computing geodesic distance; the command will read the source vertex array from this file.
   • DISTANCE_FILE: the computed distance file.
   • REFERENCE_DISTANCE_FILE: a reference distance file that stores the ground-truth distance.

License

The code is released under BSD 3-Clause License.

Contact

For any comments or questions, please contact Jiong Tao<[email protected]>, Bailin Deng <[email protected]>, or Yue Peng <[email protected]>.