SSD Surface Reconstruction (Version 3.0)
Fatih Calakli, Daniel Moreno, Gabriel Taubin
Version 3.0 is a speed optimized revision of our old (SSD 2.7) software. In this version, we use a hash octree data structure based on Morton codes which is observed to allow much more efficient access to leaf cells and their neighbors. Daniel Moreno did significant amount of work to implement and integrate the hash octree into our software. In addition, he has made several changes in the code to optimize the running time and peak memory usage.
Download ( If you download the software we will assume that you have read and agreed to the terms of the Licence Agreement. )
We provide compiler independent configuration files, CMakeLists.txt. CMake is a cross-platform, open-source build system. If you don't have CMake software installed on your system, please do so by visiting cmake.org, then follow these steps.
on Windows (Visual Studio):
1. Create a build directory
2. Open the cmake gui, choose your platform, then insert
Where is the source code : <location_of_the_source_directory>
Where to build binaries : <location_of_build_directory_you_just_created>
3. Click configure
4. Click generate
5. Open the project file ssd.sln and build ssd_recon in RELEASE mode
1. Open a terminal window
2. Create a build directory
– mkdir build ; cd build
3. Configure the package for your system:
– cmake -D CMAKE_BUILD_TYPE=RELEASE <location_of_the_source_directory>
4. Build the package:
Note: The code doesn't compile if you use the cmake gui version in Mac, so stick to the instructions given above. We will solve the issue soon.
SAMPLE DATA SET
For testing purposes, we provide one oriented points data set "angel_points.ply".
ssd_recon [options] [input file] [output file]
ssd_recon -octreeLevels 9 -weights 1 1 1 -solverTolerance 1.0e-8 -samplesPerNode 2 angel_points.ply angel_ssd_mesh.ply
This positive integer is an upper bound on the maximum depth of the octree that will be used for surface reconstruction. Running at level L corresponds to solving on a voxel grid whose resolution is no larger than 2^L x 2^L x 2^L . The default value is 8.
-weights [w0 w1 w2]
These positive floating point values specifies regularization parameters presented in the paper. We suggest using smaller lambda0 than lambda1 and lambda2, particularly when data is noisy. Otherwise, lambda2 can be set a larger value than lambda0 and lambda1 produces a smoother result (optional). The default value is w0 = 1.0, w1 = 1.0, and w2 = 1.0.
This positive floating point value specifies the tolerance of the iterative linear system solver. For noisy data, larger values in the range [1.0e-6 - 1.0e-7] is needed to provide a smoother reconstruction. Otherwise, small values in the range [1.0e-7 - 1.0e-9] can be used. The default value is 1.0e-8.
This positive integer value specifies the minimum number of sample points that should fall within an octree node. For noisy data, larger values in the range [5.0 - 10.0] is needed to provide a smoother reconstruction. Otherwise, small values in the range [1.0 - 5.0] can be used. The default value is 1.0.
This is a flag to provide description of the reconstruction process.
This string is the name of the file from which the point set will be read stored in PLY file. We can parse this PLY file in either ASCII or binary format. The file should be a PLY file (either in ascii or binary format) with groups of 6, white space delimited, numbers (x-, y-, and z-coordinates of the point's position, followed by the x-, y-, and z-coordinates of the point's normal). The number of oriented point samples should be specified in the "element vertex" field. (Look at angel_points.ply provided in the samples folder)
This string is the name of the file to which the polygon mesh will be written. The file is written in PLY format.
There are a few other optional command line arguments that we had used for debugging purposes. Feel free to experiment.
The input point cloud and our generated 3D models can be visualized using Meshlab, a free 3D mesh processing software program.