Implementation of the bounding mesh and bounding convex decomposition algorithms for single-sided mesh approximation
C++ CMake Objective-C
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
Modules
examples
images
src
thirdparty
CMakeLists.txt
LICENSE
README.md

README.md

Boundingmesh

Contributors: Andre Gaschler, Quirin Fischer, Philipp Gergen

License: 2-clause BSD, a permissive license, see LICENSE file; see License for details of linked libraries. Several features require more restrictive licenses.

Boundingmesh is a library and tool set for generating bounding meshes and bounding convex decompositions. A bounding mesh encloses a given mesh and has fewer vertices; it is single-sided approximate mesh. A bounding convex decomposition is a set of convex hulls of few vertices that enclose a given mesh.

Table of Contents

Overview

Features

  • Reduce mesh complexity restricted either by vertex count or by mesh error.
  • Import/Export various standard 3D geometry file formats: .off, .obj, .stl, .wrl.
  • Simplification by successive edge collapse.
  • Choose the direction of the changes to the mesh:
    • Only expand the model (grow outward).
    • Only shrink (grow inward).
    • Closest approximation.
  • Supports multiple algorithms to estimate the simplification error.
  • Interactive GUI

Bounding convex decomposition:

  • Generate a set of convex bodies to approximate a mesh
  • Algorithms operate on a voxel representation
  • Greedy splitting along axis-aligned planes (inspired by V-HACD)
  • Make sure the convex bodies enclose the entire original mesh (in contrast to V-HACD or other approximate convex decomposition algorithms)

Publications

Feel free to cite our publications:

Screenshots

GUI

The GUI interface.

After some simplification.

Bounding Convex Decomposition

Original model: Original teapot model

Convex hull: Convex hull

Voxel set: Voxelized model

Bounding convex decomposition: Bounding convex decomposition

Example Models

Multiple example meshes can be found in the directory /examples/. Simplified output files are named with a suffix _decimated.

The model in /examples/bunny/ originates from The Stanford 3D Scanning Repository.

Usage

License

Boundingmesh itself is under the permissive 2-clause BSD license. However, it may use the following open source software libraries, which have other licenses: Eigen, EigenQP, Coin3D, Qt, SOLID, SoQt, CGAL. Please find their source code and licenses on their respective websites. By default, EigenQP is compiled into Boundingmesh and LGPL applies unless you deactivate the option USE_EIGENQUADPROG. For instance, if you compile convex decomposition, CGAL components are linked whose GPL license applies to the rest of the code.

Installation

Dependencies

We rely on the linear algebra library Eigen for most computations.
If you perform loading of .wrl files, you also need the Coin3D toolkit.
The included GUI application also requires QT4 and SoQT4.
The modules involving convex bodies rely on QHull and CGAL.

Building from Source

We provide a CMake build file. We recommend building by

mkdir Release
cd Release
cmake ..
make

Binaries

Windows binaries may be included in the releases.

Command-Line Tools

Bounding Mesh Simplification

Bounding meshes can be generated with a command-line executable. The interface is as follows:

./boundingmesh [options] FilenameIn [FilenameOut]
  • FileIn (required): The path of the mesh file thats is to be simplified. The filename extension has to indicate one of the supported file formats .off, .obj, .stl or .wrl. Loading .wrl files requires the Coin library to be available.

  • FileOut (optional): The path where the resulting mesh will be stored. The filename extension determines the written format that has to be one of .off, .obj, .stl or .wrl. Defaults to boundingmesh_[FileIn].

  • --direction, -d: Direction of the simplification. Inward, Outward or Any.

  • --vertices, -v: The target number of vertices.

  • --error, -e: The maximum error an edge contraction may introduce.

  • --metric, -m: The metric generation algorithm to be used. Available choices: QEM_Classic, QEM_Modified, MinConstant, Diagonalization, Average.

  • --init, -i: The initialization algorithm of merge-based algorithms. Available choices: Midpoint, DistancePrimitives.

  • --faceset: If the input format is .wrl, only the faceset with this number is loaded. Supply the index as an 0-indexed integer.

  • --colored: If the output format is .wrl, disconnected submeshes will be assigned distinct colors. No arguments taken.

Bounding Convex Decomposition

Bounding convex decompositions can be generated with the following command-line executable:

./bounding-convex-decomposition [options] filename [filename_out.wrl]
  • filename: Input file may be of type .off, .obj, .stl or .wrl
  • --voxels, -x: Number of voxels to be used, a good value is 200000. Higher numbers increase quality and computation time.
  • --alpha, -a: Parameter alpha, default is 1. Increases the number of splits into convex bodies.
  • --error, -e: The maximum error that may be introduced by the bounding mesh post-processing step. Relative to the bounding box diagonal length. (Convex decomposition usually introduces a much higher error than the bounding mesh approximation.)

GUI Application

The application providing the graphical user interface is built from the files located insrc/boundingmesh-gui/. The filenames of the binaries are respectively

./boundingmesh-gui

or boundingmesh-gui.exe

Building the boundingmesh software including the GUI requires the additional dependencies Coin3, QT4, and SoQT4 to be installed.

The GUI enables easy selection of the various decimation configurations. It shows the current mesh detail and allows interactive simplification. Contraction can be started for a single edge or batches of 100 or 5000 edges at a time. Furthermore, simplification down to a certain vertex count or error limit can be started.

Linux Build

Most build dependencies are available with the most common distributions. In Ubuntu, you can install them with the command

sudo apt-get install build-essentials cmake-curses-gui libcoin60-dev libeigen3-dev libqt4-dev libqt4-opengl-dev libsoqt4-dev libqhull6 libqhull-dev

Windows Build

For Windows, we recommend to find the required libraries on the Robotics Library Website. Boundingmesh requires a subset of the dependencies the Robotics Library requires, so the build instructions from there apply to large extent.

Static Library

Boundingmesh provides a static C++ library that can be used from other software.

We provide a central header:

#include <boundingmesh.h>

The build process will create the static library libboundingmesh.a to be linked against your program.

A minimal example that loads a mesh from an .off file, simplifies it until it has 1000 or less vertices, and saves the result:

#include <boundingmesh.h> 
int main(int argc, char** argv)
{
    boundingmesh::Mesh mesh;
    mesh.loadOff("mesh.off");
    boundingmesh::Decimator decimator;
    decimator.setDirection(boundingmesh::Outward);
    decimator.setTargetVertices(1000);
    decimator.setMesh(mesh);
    std::shared_ptr<boundingmesh::Mesh> result = decimator.compute();
    result->writeOff("mesh_simplified.off");
    return 0;
};

The following documentation describes the complete library and gives brief explanations of the provided functionality.

Code Guide

The following is a rough textual description of the control flow of the programs and algorithms. Hopefully it eases the familiarization with the codebase.

Data Structures

The primitives to represent 3D geometry are declared in boundingmesh/Primitives.h. Vertices, edges and triangles all store their local neighbourhood. Referencing is always done through Indices that are valid within the mesh.
For example, assume you want to iterate through the adjacent vertices of one vertex. This is done by iterating over the adjacent triangles (their indices) and retrieving the triangle data to get the vertex indices.
Meshes can be loaded from a few file formats, which all extract vertex and triangle data encoded in some way. The vertices are added to the Mesh object and connected to form triangles. Vertex insertion is done through a proxy class VertexPositionSet which unifies vertices with equal coordinates (by checking if a vertex with the same coordinates exists).
Removal of vertices, triangles... is deferred, marking the deleted objects until a call to Mesh::cleanAndRenumber() is made.

Mesh Decimation

Mesh decimation begins with the configuration of a Decimator. Besides setting parameters to select the error metric or the stopping criterion this also includes the selection of the mesh to process. This configuration prepares the decimation queue for the main algorithm by evaluating all edges of the mesh.
The core decimation algorithm is located in Decimator::compute() and just repeatedly picks the best modification.
The modification of the mesh is done in Decimator::executeEdgeContraction(). This method uses information about which parts have to be removed or reconnected collected in Decimator::collectRemovalData().
The scoring of contractions (and simultaneously the selection of inserted points) is done in Decimator::computeEdgeContraction() by minimizing cost functions. A cost function is represented by a matrix, which is computed in the MetricGenerator class depending on the settings.

Convex Decomposition

The convex decomposition algorithm starts by rasterizing the mesh into a voxel model or VoxelSet.
The voxel set is then repeatedly divided by collecting the voxels on the two sides of a plane into different subsets (methods SegmenterSimple::compute() and VoxelSubset::partition()). The resulting subsets of voxels are then used to generate convex bodies, by collecting the points of the mesh that produced the voxels (method VoxelSubset::calculateConvexHull()).

Documentation

The whole library is contained in the namespace boundingmesh. Add it to your program by including boundingmesh.h and linking against libboundingmesh.a.
The source code is written mostly conforming to the Google C++ Style Guide.

Command-line tool

####boundingmesh-bin/main.cpp

int main(int argc, char** argv)
Reads the given arguments and adds default arguments if needed. Then the used file formats are determined and the input file loaded. The mesh is simplified and finally the results are written to the specified file.

FileFormat getFileFormat(std::string filename)
Determines the file format by extracting and matching the file extension.

enum FileFormat
Representation of a parsed file format.

option::ArgStatus checkDirection(const option::Option& option, bool msg)
option::ArgStatus checkInt(const option::Option& option, bool msg)
option::ArgStatus checkFloat(const option::Option& option, bool msg)
Callback functions for the option parser.

Graphical User Interface

boundingmesh-gui/gui.h

class MainWindow

The main window of the application, implemented using Qt. Keeps all the stateful GUI fields and declares callback functions for active components.

  • MainWindow(QWidget* parent, QApplication* app)
    Sets up all the elements of the GUI and connects the appropriate callbacks.

  • void loadView()
    Used to show a new model. Initializes the library, resets error values and displays the new mesh.

  • void reload()
    Updates current vertex count and error. Also recalculates the recommendation fields and checks button states.

class CustomLineEdit

Modified LineEdit that inserts a default value when empty.

boundingmesh-gui/ViewerMesh.h

class ViewerMesh

Converts the data of a boundingmesh::Mesh into a Coin scene graph for display.

  • void replaceByMesh(const boundingmesh::Mesh *mesh)
    Swaps the displayed mesh for a new one.

Core Library

boundingmesh/boundingmesh.h

Main header file of the library. Collects the other top-level headers.

boundingmesh/Primitives.h

Definition of the geometric primitives used to construct meshes. The linear algebra library Eigen is used to represent mathematical objects and perform computations.

The geometric primitives provide no writing access to ensure the consistency of mesh data.

typedef ... Real
Representation of a real number.

typedef ... Vector3
A three dimensional vector.

typedef ... Vector4
A 4D-vector used for homogeneous coordinates.

typedef ... Matrix44
A 4x4 matrix.

typedef ... Index
An index referencing another geometric object. The type of the object and the underlying collection always depend on the context.

class Plane

Representation of a plane in 3D space.

  • Plane()
    Create an empty plane. Normal and d are initialized to zero.

  • Plane(Vector3& normal, Real d)
    Define a plane using it's normal vector and the parameter d.

  • Plane(Vector3& normal, Vector3& point)
    Define a plane using it's normal vector and a point on the plane.

  • Plane(Vector3 point1, Vector3 point2, Vector3 point3)
    Construct a plane from three points. The orientation of the plane depends on the order of the given points; the normal is computed as the cross product of the positions of point2 and point3 relative to point 1.

  • Vector3 normal
    The plane's normal vector.

  • Real d
    Parameter for the plane equation. Points x on the plane satisfy x*normal + d = 0.

  • Real distance(const Vector& point) const
    Computes the signed distance from point to the plane.

  • Vector4 vector() const
    Returns the representation of the plane as a 4D-Vector.

  • Matrix44 distanceMatrix() const
    Returns a matrix to compute the squared distance to the plane.

class Vertex

A vertex of a mesh. Contains references to all neighbouring triangles and edges.

  • Vertex()
    Creates an empty vertex. Position initializes to zero.

  • Vertex(Vector3 position)
    Creates a vertex at position.

  • Vertex(Real x, Real y, Real z)
    Creates a vertex from three coordinates.

  • Vector3 position() const
    Getter for the 3D position.

  • unsigned int nTriangles() const
    Number of triangles containing the vertex.

  • unsigned int nEdges() const
    Number of edges containing the vertex.

  • Index triangle(unsigned int i) const
    Returns the ith triangle containing the vertex.

  • Index edge(unsigned int i) const
    Returns the ith edge containing the vertex.

class Edge

An edge of a mesh. Contains references to it's endpoints and the triangles it is part of. The references to the endpoints are always stored ordered to satisfy vertex_1 < vertex_2.

  • Edge()
    Creates an empty edge.

  • Edge(Index vertex_1, Index vertex_2)
    Creates an edge connecting vertex_1 and vertex_2.

  • unsigned int nTriangles() const
    Returns the number of triangles containing the edge.

  • bool border() const
    A flag indicating whether the edge lies on the border of the mesh.

  • Index vertex(unsigned int i) const
    Returns the ith endpoint of the edge. i may only be 0 or 1.

  • Index triangle(unsigned int i) const
    Returns the ith triangle the edge is part of. i may be 0 or 1. If border() is true, i may be only 0.

class Triangle

A triangle of a mesh. Keeps references of it's vertices and edges. Also provides the plane containing the triangle in space.

  • Triangle()
    Creates an empty triangle.

  • Triangle(Index vertex_1, Index vertex_2, Index vertex_3)
    Creates a triangle with vertices vertex_1, vertex_2, vertex_3.

  • Index vertex(unsigned int i) const
    Returns the ith vertex of the triangle. i may be 0, 1 or 2.

  • Index edge(unsigned int i) const
    Returns the ith edge of the triangle. i may be 0, 1 or 2.

  • Plane plane() const
    Returns the plane spanned by the triangles vertices.

boundingmesh/Mesh.h

class VertexPosition

Helper class to efficiently find indices of vertices that were added to the mesh.

  • VertexPosition(Vector3 position)
    Creates a vertex without specifying it's index. Used to find the indexed vertex with the same position.

  • VertexPosition(Index index, Vector3 position)
    Creates a new vertex with it's index in the mesh.

  • Index index() const
    Returns the index of the vertex.

  • friend bool operator<(const VertexPosition & lhs, const VertexPosition & rhs)
    friend bool operator>(const VertexPosition & lhs, const VertexPosition & rhs)
    Compares two vertices by their position, component-by-component.

class VertexPositionSet

Helper class to add vertices to a mesh by position. Efficiently finds the index if a vertex with the given position already exists.

  • VertexPositionSet(Mesh& mesh)
    Creates an empty set.

  • Index addVertex(Vector3 position)
    Adds a vertex with the given position and returns it's index. If there already exists a vertex on the position, it's index is returned.

class Mesh

A triangle mesh. Stores it's vertices, edges and triangles. Meshes can be changed by adding and removing vertices or triangles. Load/Save methods for some file formats are provided.

Construction
  • Mesh()
    Creates an empty mesh.

  • Mesh(const std::vector<Vector>& vertices,const std::vector<Index*>& triangles)
    Construct a mesh from a list of vertex positions and a list of index 3-tuples defining the triangles.

  • void loadOff(const std::string filename)
    void loadObj(const std::string filename)
    void loadStl(const std::string filename)
    void loadWrl(const std::string filename)
    Loads mesh data from a file. loadWrl requires the Coin library.

  • void writeOff(const std::string filename)
    void writeObj(const std::string filename)
    void writeStl(const std::string filename)
    void writeWrl(const std::string filename)
    Saves mesh data to a file.

Getter
  • unsigned int nVertices() const
    Returns the total number of vertices.

  • unsigned int nEdges() const
    Returns the total number of Edges.

  • unsigned int nTriangles() const
    Returns the total number of triangles.

  • const Vertex& vertex(Index i) const
    Returns the ith vertex of the mesh.

  • const Edge& edge(Index i) const
    Returns the ith edge of the mesh.

  • const Triangle& triangle(Index i) const
    Returns the ith triangle of the mesh.

  • bool isDirty()
    Can be used to check if the current mesh state is currently dirty because of removal actions. If it is, call void cleanAndRenumber() before iterating over geometric objects.

Modification
  • Index addVertex(const Vector3& vertex)
    Adds a new vertex to the mesh.
  • Index addTriangle(Index vertex1, Index vertex2, Index vertex3)
    Adds a new triangle to the mesh. Requires the three given indices to be valid.

Removal methods keep the mesh consistent, removing all references to the deleted object and deleting empty objects (i.e. edges whose triangles were removed). Be careful when accessing the mesh data after removal as they leave the mesh in a dirty state.

  • void removeVertex(Index vertex)
    Removes a vertex (and all triangles containing it) from the mesh.
  • void removeTriangle(Index triangle)
    Removes a triangle from the mesh.
Helper Methods (internal)

These methods are only used internally.

  • void clean()
    Resets the mesh to an empty state.

  • Index registerEdge(Index vertex1, Index vertex2, Index triangle)
    Registers a new edge within the mesh.

  • void cleanAndRenumber()
    Perfoms the deferred removal of vertices and triangles. A call to removeVertex() or removeTriangle() only marks them for later removal and removes all references to the index from the mesh. cleanAndRenumber() removes the data and calculates new indices.

  • std::vector<std::shared_ptr<Mesh> > computeSubmeshes()
    Partitions the mesh graph, every connected component is a submesh.

Static methods
  • void writeMultimeshWrl(std::vector<std::shared_ptr<Mesh> >submeshes, std::string filename, bool colored = false)
    Create a VRML file containing multiple meshes.
  • static Vector3 HSVtoRGB(Vector3 color)
    Transformation from HSV to RGB color space.
  • static std::vector<Vector3> generateColors(int n)
    Generate n distinct colors.

boundingmesh/ContractionUtils.h

Data structures used in the edge contraction algorithm.

class EdgeContraction

Stores information for the contraction of an edge.

  • EdgeContraction(Index edge, Vector new_point, Real cost, const Matrix& qem)
    Stores the contraction data for the edge with index edge: The resulting point after contraction, the cost of the introduced error and the QEM matrix of the new point.

  • Index edge() const
    Returns the index of the edge to be contracted.

  • Vector new_point() const
    Returns the position of the new point that will be inserted into the mesh instead.

  • Real cost() const
    Returns the cost of the mesh change introduced by the contraction.

  • const Matrix44& qem() const
    Returns the QEM matrix of the new point that will be used to measure mesh error.

  • friend bool operator<(const EdgeContraction & lhs, const EdgeContraction & rhs)
    friend bool operator>(const EdgeContraction & lhs, const EdgeContraction & rhs)
    Compares two contractions by their cost.

class ContractionIndex

Helper class to access contractions in the ContractionQueue by edge index.

  • ContractionIndex(Index index)
    Creates a contraction index containing only an edge index (but no iterator to a contraction). Used for finding the corresponding contraction.

  • ContractionIndex(Index index, std::set<EdgeContraction>::iterator contraction)
    Creates a new contraction index from an edge index and the iterator of the corresponding contraction.

  • std::multiset<EdgeContraction>::iterator contraction() const
    Returns the iterator of the contraction.

  • friend bool operator<(const ContractionIndex & lhs, const ContractionIndex & rhs)

  • friend bool operator>(const ContractionIndex & lhs, const ContractionIndex & rhs)
    Compares two contraction indices by their edge index.

class ContractionQueue

Container to efficiently add/remove contractions and retrieve the cheapest contraction.

  • ContractionQueue()
    Creates an empty queue.

  • unsigned int size()
    Returns the current size of the queue.

  • const EdgeContraction& first()
    Returns a reference to the first contraction in the queue (the one with the lowest cost).

  • void insert(EdgeContraction contraction)
    Inserts a contraction into the queue. Also internally creates the contraction index to access the contraction by edge index.

  • void remove(Index edge)
    Removes the contraction of the edge with index edge from the queue.

boundingmesh/Decimator.h

Contains all functionality for mesh simplification through edge contraction.

class Decimator

Performs mesh simplification by edge decimation.

Setup/Configuration
  • enum DecimationDirection
    Options to control the direction of changes to the mesh. Values are Outward, Inward and Any.

  • Decimator()
    Creates a decimator with the default configuration.

  • Decimator(int target_vertices, Real maximum_error, DecimationDirection direction)
    Creates a decimator with a custom configuration.

  • void setTargetVertices(int target_vertices)
    Changes the targeted number of vertices the output mesh should be made up of.

  • void setMaximumError(Real maximum_error)
    Sets the maximum error by which the output may differ from the input mesh.

  • void setDirection(DecimationDirection direction)
    Sets the allowed direction of changes to the mesh. The mesh may be decimated only outwards, only inwards or in any direction.

Execution
  • Mesh* compute(const Mesh& mesh)
    Simplifies the mesh according to the configuration. Returns a pointer to the resulting mesh.
Submethods (internal)
  • void initialize()
    Initializes the matrices to compute the error function and computes the initially possible contractions.
  • Matrix44 computeQEM(Index vertex)
    Computes the quadratic error matrix used to evaluate the error function near the vertex with index vertex.
  • void executeEdgeContraction(EdgeContraction contraction)
    Executes the edge contraction. Removes the referred edge, it's vertices and adjacent triangles. Then inserts the new point and connects it to the mesh.
Decimation Computation (internal)
  • EdgeContraction computeEdgeContraction(Index edge_index)
    Computes the new point and it's cost for contracting the edge with index edge_index.

  • unsigned int nSubsets(unsigned int subset_size, unsigned int total_size)
    Calculates the number of subset_size-sized subsets of a set with total_size elements (the binomial coefficient).

  • void nextSubset(std::vector<unsigned int>& indices_subset, unsigned int total_size)
    Modifies the indices in indices_subset to iterate to the next subset of a total_size-sized set.

  • bool solveConstrainedMinimisation(const Matrix44& qem, const std::vector<Plane>& constraints,
    const std::vector<unsigned int>& subset, Vector& result)
    Minimizes the error function given by the matrix qem while satisfying the constraints defined by a subset of the plane set. Stores the solution in result.

  • Vector3 minimizeSubspace(const Matrix44& qadratic_cost)

  • Vector3 minimizeSubspace(const Matrix44& qadratic_cost, Plane plane)

  • Vector3 minimizeSubspace(const Matrix44& qadratic_cost, Plane plane1, Plane plane2)

  • Vector3 minimizeSubspace(const Matrix44& qadratic_cost, Plane plane1, Plane plane2, Plane plane3)
    Minimizes the function specified by quadratic_cost with respect to 0-3 conditions. Solves in the subspace satisfying the constraints, then transforms back.

  • Vector3 minimizeLagrange(const Matrix44& qadratic_cost)

  • Vector3 minimizeLagrange(const Matrix44& qadratic_cost, Plane plane)

  • Vector3 minimizeLagrange(const Matrix44& qadratic_cost, Plane plane1, Plane plane2)

  • Vector3 minimizeLagrange(const Matrix44& qadratic_cost, Plane plane1, Plane plane2, Plane plane3)
    Directly minimizes the function given by quadratic_cost with respect to 0-3 conditions using lagrange multipliers.

boundingmesh/MetricGenerator.h

enum Metric

List of available metric generation algorithms provided by the MetricGenerator. Currently we support the following algorithms:

  • ClassicQEM
  • ModifiedQEM
  • MinimizedConstant
  • Diagonalization
  • MinimalMerge
class MetricGenerator

Produces quadratic error metrices for use by the Decimator. Allows selection of the desired metric.

Initialization
  • void setMetric(Metric metric)
    Select a generation algorithm.
  • void setInitialization(Initialization initialization)
    Select the method a merge-based algorithm should use to generate the initial metrices for vertices.
  • void setMesh(Mesh* mesh)
    Set the geometry that is being decimated.
Interface to Decimator
  • Matrix44 getErrorMetric(Index edge_index)
    Computes and returns the error metric for contracting a given edge.
  • void contractEdge(Index edge_index)
    Updates metric data to account for the contraction of an edge.
  • void cleanAndRenumber()
    Applies index changes caused by mesh cleanup to the metric data.
General Internal Methods
  • void initialize()
    Sets up the active generation algorithm for the current mesh. If a merging-based algorithm is used matrices are generated depending on the selected initialization method.
  • void shrinkIndexedArray(MatrixArray* array, std::stack<Index> deleted_indices)
    Removes all entries corresponding to indices on the stack from an array. Used for cleanup.
ClassicQEM
  • Matrix44 computeQEM(esIndex vertex_index)
ModifiedQEM
  • Matrix44 computeModifiedVertexQEM(Index vertex_index)
  • Matrix44 computeModifiedEdgeQEM(Index edge_index)
  • void contractEdgeModifiedQEM(Index edge_index)
    Update all metric data according in reaction to a contraction.
  • void collectRemovalData(...)
    Extract information about the neighbourhood of the contracted edge.
Merging Based Algorithms
  • Matrix44 computeInitialMergeMetric(Index vertex_index)
    Set up the initial error metric for a vertex
  • Matrix44 mergeMax(Matrix44 a, Matrix44 b) Merges two matrices so the result returns values greater or equal than the maximum of the evaluation of the input matrices.
  • Matrix44 mergeMatrices(Matrix44 a, Matrix44 b)
    Merges two metrices according to the selected algorithm.
  • Matrix44 mergeMinConstant(Matrix44 a, Matrix44 b)
    Implementation of MinimizedConstant merging. Addition of the matrices, minimization of the constant term.
  • Matrix44 mergeDiagonalization(Matrix44 a, Matrix44 b)
    Implementation of Diagonalization merging. Computes a diagonal bound for each matrix, merges these.
  • Matrix44 mergeAverage(Matrix44 a, Matrix44 b)
    Merges the matrices by computing their average.

boundingmesh/VoxelSet.h

enum Metric

List of possible voxel types.

  • INNER: Added to fill the inner volume.
  • SURFACE: A voxel generated by a triangle of the mesh.
class Voxel

Stores data for one voxel, currently position, type and triangles that created the voxel.

Creation
  • Voxel(Index x, Index y, Index z, VoxelType type)
    Create a voxel of certain type at some position.
  • void addTriangle(Index triangle)
    Add a triangle to a vertain voxel.
Getter Methods
  • Index x() const
  • Index y() const
  • Index z() const
    Get the coordinates of the voxel's position.
  • Index coordinate(int dimension) const
    Get the coordinate by dimension, 0->X, 1->Y, 2->Z.
  • VoxelType type() const
    Type of the voxel.
  • unsigned int nTriangles() const
    Number of triangles intersecting the voxel.
  • Index triangle(unsigned int i) const
    Returns the index of the ith triangle associated with the voxel.
class VoxelSet

A set of voxels, the equivalent of a mesh.

Creation
  • VoxelSet(std::shared_ptr<Mesh> triangle_mesh, Real voxel_size)
    Creates a voxel set by rasterizing a mesh. The grid resolution is chosen according to the mesh size (it's bounding box) and the selected voxel size.
Getters
  • unsigned int nVoxels() const
  • const Voxel& voxel(unsigned int i) const
  • const Vector3& origin() const
  • Real voxelSize() const
  • unsigned int resolution(int dimension) const
  • std::shared_ptr<Mesh> mesh() const
  • int voxelAt(unsigned int x, unsigned int y, unsigned int z)
  • Vector3 computePosition(const Voxel& voxel) const
Others
  • void addVoxel(const Voxel& voxel)
    Add a voxel to the set.
  • void writeWRL(std::string filename)
    Generate a VRML file to display the voxel set. Coloring dependent on voxel type.
Internal data structures
  • std::vector<Voxel> voxels_
    Set of all filled voxels.
  • std::vector<int> grid_
    Field of size resolution(0)*resolution(1)*resolution(2), storing the index of the voxel for a certain position.

boundingmesh/SegmenterSimple.h

Classes for the segmentation of a model into disjoint parts, especially for convex decomposition.

class Split

A splitting plane, currently axis aligned and at a voxel boundary.

  • Split(int dimension, int index)

  • int dimension
    Dimension perpendicular to the split plane.

  • int index
    Split between index and index+1.

  • bool test(const Voxel& voxel)
    Classification of a voxel using the split. True if it's on the upper side of the split (greater coordinate than the split).

class AppliedSplit

Information attached to a set after splitting. Stores the splitting plane and the side of this part.

  • AppliedSplit(Split& split, bool direction)

  • Split split
    The corresponding split.

  • bool direction
    Indicator which side of the plane was chosen.

  • Plane getPlane(const VoxelSet& voxel_set)
    Mathematical plane of, oriented so that this subset lies on the front (distance > 0).

  • static int mergeSplits(const AppliedSplit& a, const AppliedSplit& b)
    Evaluation whether two splits can be merged (because one is a subset of the other).

class VoxelSubset

A subset of a voxelized mesh. Keeps a list of the voxels part of this subset and information about the splits made. A subset generates one convex body.

Creation
  • VoxelSubset(std::shared_ptr<VoxelSet> voxels)
    Creates an empty subset linked to a voxel set.

  • VoxelSubset(const VoxelSubset& subset, bool empty = false)
    Creates a new subset by copying an existing one. Refine a subset by copying it, but removing the voxels(with set empty flag). Then add the voxels that fit your criteria.

  • void addVoxel(Index index)
    Add a voxel to the subset.

  • void addSplit(AppliedSplit split)
    Add a split that clips the subset.

  • void calculateConvexHull()
    Compute the convex hull of the triangles associated with the subset. Clips triangles against the splits.

Further partition
  • std::vector<VoxelSubset> partition(Split split)
    Partition into two sets according to a split.
  • void setFinal()
    Sets the flag to stop partitioning this set.
Getter
  • Real convexVolume() const
    Volume of the associated convex body.

  • Real volume() const
    Volume of the geometry of the subset, approximated by the filled voxels.

  • std::shared_ptr<Mesh> getConvexHull()
    Get the mesh of the surface of the subsets convex hull.

  • bool isFinal()
    Check whether the subset should be further split.

class SegmenterSimple
  • SegmenterSimple(Real voxel_size) Create a segmenter, voxel_size will be used to rasterize the mesh.
Configuration
  • void setMaxPasses(int passes)
    Set the number of passes (iterations of splitting the subsets).

  • void setMinVolumeGain(Real min_gain)
    Set the minimal improvement of the cost function that causes a partitioning.

  • void setMesh(std::shared_ptr <Mesh> mesh)
    Select the model to segment/decompose.

Main algorithm
  • void compute()
    Compute the segmentation of the mesh
Results
  • std::vector< std::shared_ptr<Mesh> > getSegmentation()
    Get the convex meshes of the segmentation.

####### Internal methods

  • std::vector<Split> generateSplits(const VoxelSubset& subset)
    Compute all possible splits that could partition a certain subset. They will will be evaluated and the one with the best rating is executed.
  • Real evaluatePartition(const VoxelSubset& previous_set, const std::vector<VoxelSubset>& partition)
    Compute heuristic evaluation of previous set into a partition.
  • Real evaluateSubset(const VoxelSubset& subset)
    Compute evaluation of a subset (currently normalized to 1).