Symmetric Poisson Reconstruction (Version 1.00)

This software supports reconstruction of surfaces from unoriented points.

LINKS

Papers:

[Kohlbrenner, Liu, Alexa, and Kazhdan, 2025]

Executables:

Win64

Source Code:

ZIP

GitHub

EXECUTABLES

--in <input points>: This string is the name of the file from which the point set will be read.
If the file extension is .ply, the file should be in PLY format, giving the list of oriented vertices with the x-, y-, and z-coordinates of the positions encoded by the properties x, y, and z.
Otherwise, the file should be an ascii file with groups of 3, white space delimited, numbers: the x-, y-, and z-coordinates of the point's position. (No information about the number of oriented point samples should be specified.)
[--out <output mesh/implicit-function>]: This string is the name of the file to which the output will be written.

If the file extension is .ply, the output will be a polygon/triangle mesh (output in PLY format).

If the file extesnion is .grid, the output will be regular grid sampling the implicit function. The file will consist of seven header lines in ASCII describing the contents, following by the grid values in binary.

The header looks like:
G3 1 DOUBLE <RES_X> <RES_Y> <RES_Z> <M_00> <M_01> <M_02> <M_03> <M_10> <M_11> <M_12> <M_13> <M_20> <M_21> <M_22> <M_23> <M_30> <M_31> <M_22> <M_33>
The first two lines describe the contents of the file -- a 3D grid with a double-precision floating point value per cell.
The next line gives the resolution of the grid in x-, y-, and z-directions.
The following four lines give the 4x4 coefficients of the homogenous transformation <M> taking grid-space coordinates to world-coordinates.
The next 8 x <RES_X> x <RES_Y> x <RES_Z> bytes correspond to the (double-precision) floating point values of the implicit function.
[--depth <reconstruction depth>]: This integer is the depth of the tree that will be used for surface reconstruction. Running at depth d corresponds to solving on a voxel grid whose resolution is 2^d x 2^d x 2^d. Note that since the reconstructor adapts the octree to the sampling density.
The default value for this parameter is 7.
[--sWeight <screening weight>]: This floating point value specifies the weight given to having the implicit function evaluate to zero at the samples.
The default value is 50,000.
[--dWeight <boundary Dirichlet weight>]: This floating point value specifies the weight given to having the implicit function be constant on the boundary of the unit-cube.
The default value is 100.
[--scale <scale factor>]: This floating point value specifies the ratio between the diameter of the cube used for reconstruction and the diameter of the samples' bounding cube.
The default value is 1.25.
[--coarseIiters <Gauss-Seidel iterations at coarsest level>]: This integer value specifies the number of Gauss-Seidel relaxations to be performed at the coarsest level of the hierarchy.
The default value for this parameter is 512.
[--iters <Gauss-Seidel iterations at finest level level>]: This integer value specifies the number of Gauss-Seidel relaxations to be performed at the finest level of the hierarchy.
The default value for this parameter is 10.
[--iMult <Gauss-Seidel iteration multiplier>]: This floating point value how the multiplicative factor for the number of Gauss Seidel iterations to be performed at progressively coarser levels (excluding the coarsest level).
The default value for this parameter is 2.0.
[--nn <number of nearest neighbors>]: This integer value specifies the number of nearest neighobrs that are to be used for estimating the local covariance.
The default value for this parameter is 20.
[--density]: Enabling this flag tells the reconstructor to output a triangle mesh (rather than a polygon mesh).
[--verbose <verbosity>]: This integer variable specifies the level of verbosity output to the standard output (with larger values corresponding to more verbose output).
Note: The calculation of verbose output can noticeably affect run-time performance.

USAGE EXAMPLES (WITH SAMPLE DATA)

SymmetricPoissonRecon

For testing purposes, a point-set sampling the Stanford Bunny is provided. (Data sourced from the Stanford Scanning Repository.) The point-set consists of roughly 360K points, written in ASCII format with three floating point values per line.
The surface of the model can be reconstructed by calling symmetric Poisson reconstructor:

% SymmetricPoissonRecon --in bunny_raw.xyz --out bunny.ply

HISTORY OF CHANGES

Version 1:

Initial source-code.

HOME