repository·master·Indexed 23 days ago
https://github.com/mkazhdan/poissonreconA high-performance library for general adaptive finite-element (FEM) systems, evolved from Poisson Surface Reconstruction. It supports spaces of arbitrary dimension, discretization using finite-elements of arbitrary degree, and both point-wise and integrated constraints. The package includes tools such as PoissonRecon for triangle mesh reconstruction from oriented 3D points, SSDRecon for surface mesh reconstruction via Smooth Signed Distance functions, and PoissonReconServer/Client for distributed reconstruction.
The poissonrecon codebase is an evolution of the Poisson Surface Reconstruction code, designed to support general adaptive finite-element (FEM) systems. It is capable of handling:
PoissonRecon and SSDRecon support processing 2D point sets. The 2D implementations of PoissonRecon, SSDRecon, and AdaptiveTreeVisualization also support outputting to .jpg and .png image files.When using libjpeg-turbo, observe these universal constraints:
libjpeg-turbo uses two compatible BSD-style licenses depending on which part of the library you are using:
libjpeg API library and associated programs (including inherited code and modifications). Note that while SIMD source code bears the zlib License, its terms are subsumed by the IJG License in the context of the overall library.TurboJPEG API library, associated programs, and the build system. Because the TurboJPEG API wraps the libjpeg API, both the IJG License and the Modified BSD License apply when using the TurboJPEG API.If you are distributing only libjpeg-turbo binaries (without source) or an application that statically links with libjpeg-turbo, you must fulfill these documentation requirements:
This software is based in part on the work of the Independent JPEG Group.To use the full feature set of the library, ensure the following dependencies are met:
zlib, png, and jpg libraries. The Makefile expects header files in /usr/local/include/ and library files in /usr/local/lib/.Use PointInterpolant to fit a function to a set of samples, and AdaptiveTreeVisualization to evaluate that function at specific points.
PointInterpolant with --inValues (samples) and --tree (output tree file).AdaptiveTreeVisualization with --in (the tree file) and --samples (evaluation points).The reconstruction code can be interfaced via Reconstructors.h and Extrapolator.h. To use the library, you must choose a finite element type and define three specific data streams by overriding virtual classes.
Reconstruction requires template parameters for the floating point type (Real, typically float) and the dimension (Dim, fixed at 3).
FEMSig: Describes the 1D finite element type. It is a composite of the Degree (integer) and BoundaryType (one of BOUNDARY_FREE, BOUNDARY_DIRICHLET, or BOUNDARY_NEUMANN defined in BSplineData.h).static const unsigned int FEMSig = FEMDegreeAndBType< Degree , BoundaryType >::Signature;FEMSigs: Describes the tensor-product finite element type (typically isotropic).using FEMSigs = IsotropicUIntPack< Dim , FEMSig >; You must implement the following three stream classes by overriding their virtual methods:
Derive from InputSampleStream< Real , Dim >. Overrides:
void reset(): Resets the stream to the start (required for the two-pass reconstruction process).bool read( Point< Real , Dim > &p , Point< Real , Dim > &n ): Reads the next position/normal pair. Returns true if successful, false at end-of-stream.Derive from OutputPolygonStream. Overrides:
size_t size(): Returns the number of polygons written.size_t write( const std::vector< node_index_type > &polygon ): Writes a polygon as a vector of integral indices. Returns the index of the written polygon.Derive from OutputVertexStream< Real , Dim >. Overrides:
size_t size(): Returns the number of vertices written.size_t write( Point< Real , Dim > p , Point< Real , Dim > g , Real w ): Writes vertex position p, gradient g, and density weight w. Returns the index of the written vertex.For maximum efficiency, the code is optimized to compile for a specific FEM degree and boundary condition. By default, DEFAULT_FEM_DEGREE is set to 1 and DEFAULT_FEM_BOUNDARY is set to Neumann in PoissonRecon.h.
To support varying FEM degrees and boundary conditions at runtime, you must #undef the FAST_COMPILE pre-processor definition in PreProcess.h. Note that this will significantly increase compilation time.
If you are distributing a modified version of the libjpeg-turbo source code, you must adhere to the following requirements:
README.ijg file in your distribution and must not alter any copyright or license text within it.The PoissonRecon tool reconstructs a triangle mesh from a set of oriented 3D points by solving a Poisson system.
--inx, y, z and normal properties nx, ny, nz.x, y, z (position) followed by nx, ny, nz (normal).x, y, z (position) followed by nx, ny, nz (normal).--in <input points>: Path to the input point set.--out <output triangle mesh>: Path to the output PLY file.--envelope <constraint envelope>: (Optional) A water-tight PLY triangle mesh used as a constraint. Normals must point outside the reconstructed surface.--depth <reconstruction depth>: Maximum octree depth (default: 8).--width <finest cell width>: Target width of the finest level octree cells (ignored if --depth is used).--scale <scale factor>: Ratio between the reconstruction cube diameter and the samples' bounding cube diameter (default: 1.1).--samplesPerNode <minimum number of samples>: Minimum samples per octree node. Use 1.0 - 5.0 for noise-free samples, or 15.0 - 20.0 for noisy samples (default: 1.5).--colors: Enables color interpolation if input points (ASCII/binary) contain color values (9 floats per sample instead of 6).BIG_DATA in the file PreProcessor.h.