manolas
/
vcglib
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Refactoring of the voronoi processing framework 

factored out the point sampled distance computation and the approximate
skeleton
This commit is contained in:
Paolo Cignoni 2016-12-20 09:12:46 +01:00
parent 96087ff8e5
commit 611341b754
3 changed files with 257 additions and 159 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

132
vcg/complex/algorithms/skeleton.h Normal file
View File

@ -0,0 +1,132 @@
/****************************************************************************
* VCGLib o o *
* Visual and Computer Graphics Library o o *
* _ O _ *
* Copyright(C) 2004-2016 \/)\/ *
* Visual Computing Lab /\/| *
* ISTI - Italian National Research Council | *
* \ *
* All rights reserved. *
* *
* This program is free software; you can redistribute it and/or modify *
* it under the terms of the GNU General Public License as published by *
* the Free Software Foundation; either version 2 of the License, or *
* (at your option) any later version. *
* *
* This program is distributed in the hope that it will be useful, *
* but WITHOUT ANY WARRANTY; without even the implied warranty of *
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the *
* GNU General Public License (http://www.gnu.org/licenses/gpl.txt) *
* for more details. *
* *
****************************************************************************/
#ifndef VCG__SKELETON_H
#define VCG__SKELETON_H
#include<vcg/complex/algorithms/voronoi_volume_sampling.h>
namespace vcg
{
namespace tri
{
template <class MeshType>
class SampledSkeleton
{
public:
typedef typename MeshType::ScalarType ScalarType;
typedef typename MeshType::BoxType BoxType;
typedef typename MeshType::VertexIterator VertexIterator;
typedef typename MeshType::VertexPointer VertexPointer;
typedef typename MeshType::CoordType CoordType;
typedef typename MeshType::FacePointer FacePointer;
typedef typename MeshType::FaceType FaceType;
typedef VoronoiVolumeSampling<MeshType> VoronoiVolumeSamplingType;
SampledSkeleton(VoronoiVolumeSamplingType &_vvs):vvs(_vvs){}
VoronoiVolumeSamplingType &vvs;
/**
* @brief Compute an evaulation of the thickness as distance from the medial axis.
* It starts from a montecarlo volume sampling and try to search for the samples that can be part of the medial axis.
* It use a sampled representation of the surface. A volume sample is considered part
* of the medial axis if there are at least two points that are (almost) the same minimal distance to that point.
*
*
*/
void ThicknessEvaluator(float distThr, int smoothSize, int smoothIter, MeshType *skelM=0)
{
tri::UpdateQuality<MeshType>::VertexConstant(vvs.psd.poissonSurfaceMesh,0);
std::vector<VertexPointer> medialSrc(vvs.psd.poissonSurfaceMesh.vert.size(),0);
for(VertexIterator vi=vvs.montecarloVolumeMesh.vert.begin(); vi!=vvs.montecarloVolumeMesh.vert.end(); ++vi)
{
unsigned int ind;
ScalarType sqdist;
this->vvs.psd.surfTree->doQueryClosest(vi->P(),ind,sqdist);
VertexPointer vp = &vvs.psd.poissonSurfaceMesh.vert[ind];
ScalarType minDist = math::Sqrt(sqdist);
if(vp->Q() < minDist)
{
std::vector<unsigned int> indVec;
std::vector<ScalarType> sqDistVec;
this->vvs.psd.surfTree->doQueryDist( vi->P(), minDist*distThr,indVec,sqDistVec);
if(indVec.size()>1)
{
for(size_t i=0;i<indVec.size();++i)
{
VertexPointer vp = &vvs.psd.poissonSurfaceMesh.vert[indVec[i]];
//ScalarType dist = math::Sqrt(sqDistVec[i]);
if(vp->Q() < minDist) {
vp->Q()=minDist;
medialSrc[indVec[i]]=&*vi;
}
}
}
}
}
// Now collect the vertexes of the volume mesh that are on the medial surface
if(skelM)
{
tri::UpdateFlags<MeshType>::VertexClearV(vvs.montecarloVolumeMesh);
for(size_t i=0;i<medialSrc.size();++i)
medialSrc[i]->SetV();
for(VertexIterator vi=vvs.montecarloVolumeMesh.vert.begin(); vi!=vvs.montecarloVolumeMesh.vert.end(); ++vi)
if(vi->IsV()) tri::Allocator<MeshType>::AddVertex(*skelM,vi->P());
printf("Generated a medial surf of %i vertexes\n",skelM->vn);
}
tri::Smooth<MeshType>::PointCloudQualityMedian(vvs.psd.poissonSurfaceMesh);
tri::Smooth<MeshType>::PointCloudQualityAverage(vvs.psd.poissonSurfaceMesh,smoothSize,smoothIter);
tri::UpdateColor<MeshType>::PerVertexQualityRamp(vvs.psd.poissonSurfaceMesh);
tri::RedetailSampler<MeshType> rs;
rs.init(&vvs.psd.poissonSurfaceMesh);
rs.dist_upper_bound = vvs.psd.poissonSurfaceMesh.bbox.Diag()*0.05 ;
rs.qualityFlag = true;
tri::SurfaceSampling<MeshType, RedetailSampler<MeshType> >::VertexUniform(vvs.baseMesh, rs, vvs.baseMesh.vn, false);
}
void RefineSkeletonVolume(MeshType &skelMesh)
{
CoordType closestP;
int trialNum=0;
for(int i=0;i<skelMesh.vn;++i)
{
CoordType point = math::GeneratePointInBox3Uniform(vvs.rng,vvs.baseMesh.bbox);
trialNum++;
ScalarType d = this->DistanceFromSurface(point, closestP);
if(d<0){
vcg::tri::Allocator<MeshType>::AddVertex(vvs.montecarloVolumeMesh,point);
vvs.montecarloVolumeMesh.vert.back().Q() = fabs(d);
}
}
}
}; // end class
} // end namespace vcg
} // end namespace vcg
#endif // VCG__SKELETON_H

4
vcg/complex/algorithms/voronoi_processing.h
View File

@ -79,7 +79,7 @@ struct VoronoiProcessingParameter
bool relaxOnlyConstrainedFlag;
bool preserveFixedSeed; /// If true the 'fixed' seeds are not moved during relaxation.
/// \see FixVertexVector function to see how to fix a set of seeds.
/// \see MarkVertexVectorAsFixed function to see how to fix a set of seeds.
float refinementRatio; /// It defines how much the input mesh has to be refined in order to have a supporting
/// triangulation that is dense enough to well approximate the voronoi diagram.
@ -1195,7 +1195,7 @@ static void PruneSeedByRegionArea(std::vector<VertexType *> &seedVec,
/// Vertex pointers must belong to the mesh.
/// The framework use a boolean attribute called "fixed" to store this info.
///
static void FixVertexVector(MeshType &m, std::vector<VertexType *> &vertToFixVec)
static void MarkVertexVectorAsFixed(MeshType &m, std::vector<VertexType *> &vertToFixVec)
{
typename MeshType::template PerVertexAttributeHandle<bool> fixed;
fixed = tri::Allocator<MeshType>:: template GetPerVertexAttribute<bool> (m,"fixed");

274
vcg/complex/algorithms/voronoi_volume_sampling.h
View File

@ -33,6 +33,103 @@ namespace vcg
namespace tri
{
template <class MeshType>
class PointSampledDistance
{
public:
typedef typename MeshType::ScalarType ScalarType;
typedef typename MeshType::BoxType BoxType;
typedef typename MeshType::VertexIterator VertexIterator;
typedef typename MeshType::VertexPointer VertexPointer;
typedef typename MeshType::CoordType CoordType;
typedef typename MeshType::FacePointer FacePointer;
typedef typename MeshType::FaceType FaceType;
typedef typename vcg::GridStaticPtr<typename MeshType::FaceType, ScalarType> GridType;
typedef SimpleVolume<SimpleVoxel<ScalarType> > VVSVolume;
typedef typename vcg::tri::TrivialWalker<MeshType,VVSVolume> VVSWalker;
typedef typename vcg::tri::MarchingCubes<MeshType, VVSWalker> VVSMarchingCubes;
PointSampledDistance(MeshType &_baseMesh)
:surfTree(0),baseMesh(_baseMesh) {}
typename KdTree<ScalarType>::PriorityQueue pq;
GridType surfGrid; // used for fast inside query
typedef FaceTmark<MeshType> MarkerFace;
MarkerFace mf;
vcg::face::PointDistanceBaseFunctor<ScalarType> PDistFunct;
KdTree<ScalarType> *surfTree; // used for fast inside query
MeshType &baseMesh;
MeshType poissonSurfaceMesh;
ScalarType poissonRadiusSurface;
void Init(ScalarType _poissonRadiusSurface=0)
{
MeshType montecarloSurfaceMesh;
if(_poissonRadiusSurface==0) poissonRadiusSurface = baseMesh.bbox.Diag()/50.0f;
else poissonRadiusSurface = _poissonRadiusSurface;
ScalarType meshArea = Stat<MeshType>::ComputeMeshArea(baseMesh);
int MontecarloSurfSampleNum = 10 * meshArea / (poissonRadiusSurface*poissonRadiusSurface);
tri::MeshSampler<MeshType> sampler(montecarloSurfaceMesh);
tri::SurfaceSampling<MeshType,tri::MeshSampler<MeshType> >::Montecarlo(baseMesh, sampler, MontecarloSurfSampleNum);
montecarloSurfaceMesh.bbox = baseMesh.bbox; // we want the same bounding box
poissonSurfaceMesh.Clear();
tri::MeshSampler<MeshType> mps(poissonSurfaceMesh);
typename tri::SurfaceSampling<MeshType,tri::MeshSampler<MeshType> >::PoissonDiskParam pp;
pp.geodesicDistanceFlag=false;
tri::SurfaceSampling<MeshType,tri::MeshSampler<MeshType> >::PoissonDiskPruning(mps, montecarloSurfaceMesh, poissonRadiusSurface,pp);
vcg::tri::UpdateBounding<MeshType>::Box(poissonSurfaceMesh);
printf("Surface Sampling radius %f - montecarlo %ivn - Poisson %ivn\n",poissonRadiusSurface,montecarloSurfaceMesh.vn,poissonSurfaceMesh.vn);
VertexConstDataWrapper<MeshType> ww(poissonSurfaceMesh);
if(surfTree) delete surfTree;
surfTree = new KdTree<ScalarType>(ww);
surfGrid.SetWithRadius(baseMesh.face.begin(),baseMesh.face.end(),poissonRadiusSurface);
mf.SetMesh(&baseMesh);
}
// Compute the signed distance from the surface exploting both a kdtree and a ugrid
// for a query point p first we use the kdtree with a good poisson sampling of the surface;
// to get the nearest point on the surface, then if the point is far from the surface we can use the point point distance, while if it is near (e.g. less than 3*poisson radius) we rely on point face distance with a grid.
ScalarType DistanceFromSurface(const CoordType &q, CoordType &closestP)
{
ScalarType squaredDist;
unsigned int ind;
surfTree->doQueryClosest(q,ind,squaredDist);
ScalarType dist = sqrt(squaredDist);
if( dist > 3.0f*poissonRadiusSurface)
{
// CoordType dir = surfTree->getNeighbor(0) - p;
CoordType dir = this->poissonSurfaceMesh.vert[ind].P() - q;
const CoordType &surfN = this->poissonSurfaceMesh.vert[ind].N();
if(dir* surfN > 0) dist= -dist;
closestP=this->poissonSurfaceMesh.vert[ind].P();
return dist;
}
ScalarType _maxDist = this->poissonRadiusSurface*3.0f;
dist=_maxDist;
FacePointer f=surfGrid.GetClosest(PDistFunct,mf,q,_maxDist,dist,closestP);
assert(f);
assert (dist >=0);
CoordType dir = closestP - q;
if(dir*f->cN() > 0) dist = -dist;
return dist;
}
};
/** Compute a well distributed set of samples (seeds) inside a watertight mesh.
*
* The main idea is that we have start from a poisson disk distribution and we improve it using Lloyd relaxation.
* To make things simpler and more controllable we estabilish since the beginning a Domain where we can choose the points.
*
*/
template< class MeshType>
class VoronoiVolumeSampling
{
@ -69,33 +166,26 @@ public:
};
VoronoiVolumeSampling(MeshType &_baseMesh)
:surfTree(0),seedTree(0),baseMesh(_baseMesh),cb(0),restrictedRelaxationFlag(false)
:seedTree(0),baseMesh(_baseMesh),cb(0),restrictedRelaxationFlag(false),psd(_baseMesh)
{
tri::RequirePerFaceMark(baseMesh);
tri::UpdateBounding<MeshType>::Box(baseMesh);
tri::UpdateNormal<MeshType>::PerFaceNormalized(baseMesh);
}
KdTree<ScalarType> *surfTree; // used for fast inside query
KdTree<ScalarType> *seedTree; // used to accumulate barycenter in relaxation
KdTree<ScalarType> *seedDomainTree; // used to accumulate barycenter in relaxation
typename KdTree<ScalarType>::PriorityQueue pq;
GridType surfGrid; // used for fast inside query
typedef FaceTmark<MeshType> MarkerFace;
MarkerFace mf;
vcg::face::PointDistanceBaseFunctor<ScalarType> PDistFunct;
MeshType &baseMesh;
MeshType &baseMesh; // The base mesh for which we compute all
MeshType seedMesh;
MeshType poissonSurfaceMesh;
ScalarType poissonRadiusSurface;
MeshType montecarloVolumeMesh; // we use this mesh as volume evaluator
MeshType montecarloVolumeMesh; // we use this mesh as volume evaluator and to choose
MeshType seedDomainMesh; // where we choose the seeds (by default is the montecarlo volume mesh)
vcg::CallBackPos *cb;
math::MarsenneTwisterRNG rng;
bool restrictedRelaxationFlag;
PointSampledDistance<MeshType> psd;
// Build up the needed structure for efficient point in mesh search.
// It uses a poisson disk sampling of the surface plus a
@ -103,62 +193,11 @@ public:
// It initializes the surfGrid, surfTree and poissonSurfaceMesh members
void Init(ScalarType _poissonRadiusSurface=0)
{
MeshType montecarloSurfaceMesh;
if(_poissonRadiusSurface==0) poissonRadiusSurface = baseMesh.bbox.Diag()/50.0f;
else poissonRadiusSurface = _poissonRadiusSurface;
ScalarType meshArea = Stat<MeshType>::ComputeMeshArea(baseMesh);
int MontecarloSurfSampleNum = 10 * meshArea / (poissonRadiusSurface*poissonRadiusSurface);
tri::MeshSampler<MeshType> sampler(montecarloSurfaceMesh);
psd.Init(_poissonRadiusSurface);
tri::SurfaceSampling<MeshType,tri::MeshSampler<MeshType> >::SamplingRandomGenerator()=rng;
tri::SurfaceSampling<MeshType,tri::MeshSampler<MeshType> >::Montecarlo(baseMesh, sampler, MontecarloSurfSampleNum);
montecarloSurfaceMesh.bbox = baseMesh.bbox; // we want the same bounding box
poissonSurfaceMesh.Clear();
tri::MeshSampler<MeshType> mps(poissonSurfaceMesh);
typename tri::SurfaceSampling<MeshType,tri::MeshSampler<MeshType> >::PoissonDiskParam pp;
pp.geodesicDistanceFlag=false;
tri::SurfaceSampling<MeshType,tri::MeshSampler<MeshType> >::PoissonDiskPruning(mps, montecarloSurfaceMesh, poissonRadiusSurface,pp);
vcg::tri::UpdateBounding<MeshType>::Box(poissonSurfaceMesh);
printf("Surface Sampling radius %f - montecarlo %ivn - Poisson %ivn\n",poissonRadiusSurface,montecarloSurfaceMesh.vn,poissonSurfaceMesh.vn);
VertexConstDataWrapper<MeshType> ww(poissonSurfaceMesh);
if(surfTree) delete surfTree;
surfTree = new KdTree<ScalarType>(ww);
surfGrid.SetWithRadius(baseMesh.face.begin(),baseMesh.face.end(),poissonRadiusSurface);
mf.SetMesh(&baseMesh);
}
// Compute the signed distance from the surface exploting both a kdtree and a ugrid
// for a query point p first we use the kdtree with a good poisson sampling of the surface;
// to get the nearest point on the surface, then if the point is far from the surface we can use the point point distance, while if it is near (e.g. less than 3*poisson radius) we rely on point face distance with a grid.
ScalarType DistanceFromSurface(const CoordType &q, CoordType &closestP)
{
ScalarType squaredDist;
unsigned int ind;
surfTree->doQueryClosest(q,ind,squaredDist);
ScalarType dist = sqrt(squaredDist);
if( dist > 3.0f*poissonRadiusSurface)
{
// CoordType dir = surfTree->getNeighbor(0) - p;
CoordType dir = this->poissonSurfaceMesh.vert[ind].P() - q;
const CoordType &surfN = this->poissonSurfaceMesh.vert[ind].N();
if(dir* surfN > 0) dist= -dist;
closestP=this->poissonSurfaceMesh.vert[ind].P();
return dist;
}
ScalarType _maxDist = this->poissonRadiusSurface*3.0f;
dist=_maxDist;
FacePointer f=surfGrid.GetClosest(PDistFunct,mf,q,_maxDist,dist,closestP);
assert(f);
assert (dist >=0);
CoordType dir = closestP - q;
if(dir*f->cN() > 0) dist = -dist;
return dist;
}
ScalarType DistanceFromVoronoiSeed(const CoordType &p_point)
@ -389,7 +428,7 @@ void QuadricRelaxVoronoiSamples(int relaxStep)
ScalarType ImplicitFunction(const CoordType &p, const Param &pp)
{
CoordType closest;
ScalarType surfDist = this->DistanceFromSurface(p,closest);
ScalarType surfDist = this->psd.DistanceFromSurface(p,closest);
ScalarType elemDist;
switch(pp.elemType)
@ -543,86 +582,16 @@ void OptimizeIsosurf(MeshType &m, const Param &pp)
printf("Optimize Isosurf performed %i edge flip in %5.2f s\n",flipCnt,float(t1-t0)/CLOCKS_PER_SEC);
}
/**
* @brief Compute an evaulation of the thickness as distance from the medial axis.
* It starts from a montecarlo volume sampling and try to search for the samples that can be part of the medial axis.
* It use a sampled representation of the surface. A volume sample is considered part
* of the medial axis if there are at least two points that are (almost) the same minimal distance to that point.
*
*
/** Given a surface sampling it adds to the montecarloVolumeMesh, a number of near surface samples.
* For each surface it try to add a sample generated as a point in the half ball of <radius> centered on the sample.
*/
void ThicknessEvaluator(float distThr, int smoothSize, int smoothIter, MeshType *skelM=0)
{
tri::UpdateQuality<MeshType>::VertexConstant(poissonSurfaceMesh,0);
std::vector<VertexPointer> medialSrc(poissonSurfaceMesh.vert.size(),0);
for(VertexIterator vi=montecarloVolumeMesh.vert.begin(); vi!=montecarloVolumeMesh.vert.end(); ++vi)
{
unsigned int ind;
ScalarType sqdist;
this->surfTree->doQueryClosest(vi->P(),ind,sqdist);
VertexPointer vp = &poissonSurfaceMesh.vert[ind];
ScalarType minDist = math::Sqrt(sqdist);
if(vp->Q() < minDist)
{
std::vector<unsigned int> indVec;
std::vector<ScalarType> sqDistVec;
this->surfTree->doQueryDist( vi->P(), minDist*distThr,indVec,sqDistVec);
if(indVec.size()>1)
void RefineMontecarloVolumeSamplingNearSurface(MeshType &surfaceSamplingMesh, ScalarType radius, int perSampleNum)
{
for(size_t i=0;i<indVec.size();++i)
{
VertexPointer vp = &poissonSurfaceMesh.vert[indVec[i]];
//ScalarType dist = math::Sqrt(sqDistVec[i]);
if(vp->Q() < minDist) {
vp->Q()=minDist;
medialSrc[indVec[i]]=&*vi;
}
}
}
}
}
// Now collect the vertexes of the volume mesh that are on the medial surface
if(skelM)
{
tri::UpdateFlags<MeshType>::VertexClearV(montecarloVolumeMesh);
for(size_t i=0;i<medialSrc.size();++i)
medialSrc[i]->SetV();
for(VertexIterator vi=montecarloVolumeMesh.vert.begin(); vi!=montecarloVolumeMesh.vert.end(); ++vi)
if(vi->IsV()) tri::Allocator<MeshType>::AddVertex(*skelM,vi->P());
printf("Generated a medial surf of %i vertexes\n",skelM->vn);
}
tri::Smooth<MeshType>::PointCloudQualityMedian(poissonSurfaceMesh);
tri::Smooth<MeshType>::PointCloudQualityAverage(poissonSurfaceMesh,smoothSize,smoothIter);
tri::UpdateColor<MeshType>::PerVertexQualityRamp(poissonSurfaceMesh);
tri::RedetailSampler<MeshType> rs;
rs.init(&poissonSurfaceMesh);
rs.dist_upper_bound = poissonSurfaceMesh.bbox.Diag()*0.05 ;
rs.qualityFlag = true;
tri::SurfaceSampling<MeshType, RedetailSampler<MeshType> >::VertexUniform(baseMesh, rs, baseMesh.vn, false);
}
void RefineSkeletonVolume(MeshType &skelMesh)
{
CoordType closestP;
int trialNum=0;
for(int i=0;i<skelMesh.vn;++i)
{
CoordType point = math::GeneratePointInBox3Uniform(rng,baseMesh.bbox);
trialNum++;
ScalarType d = this->DistanceFromSurface(point, closestP);
if(d<0){
vcg::tri::Allocator<MeshType>::AddVertex(montecarloVolumeMesh,point);
montecarloVolumeMesh.vert.back().Q() = fabs(d);
}
}
}
void BuildMontecarloSampling(int montecarloSampleNum)
void BuildMontecarloVolumeSampling(int montecarloSampleNum)
{
montecarloVolumeMesh.Clear();
@ -632,7 +601,7 @@ void OptimizeIsosurf(MeshType &m, const Param &pp)
{
CoordType point = math::GeneratePointInBox3Uniform(rng,baseMesh.bbox);
trialNum++;
ScalarType d = this->DistanceFromSurface(point,closest);
ScalarType d = this->psd.DistanceFromSurface(point,closest);
if(d<0){
vcg::tri::Allocator<MeshType>::AddVertex(montecarloVolumeMesh,point);
montecarloVolumeMesh.vert.back().Q() = fabs(d);
@ -647,26 +616,23 @@ void OptimizeIsosurf(MeshType &m, const Param &pp)
/*
* Function: BuildVolumeSampling
* ----------------------------
* Build a Poisson-Disk Point cloud that cover all the space of the original mesh m
* Build and prepare the seed set.
* This is the starting point for the subsequent relaxation calls.
* You can insert some initial seeds into the seed set and they will be preserved
*
*
*/
void BuildVolumeSampling(int montecarloSampleNum, int poissonSampleNum, ScalarType &poissonRadius, int randSeed)
void BuildVolumeSampling(int montecarloSampleNum, ScalarType &poissonRadius, int randSeed)
{
if(montecarloSampleNum >0)
this->BuildMontecarloSampling(montecarloSampleNum);
if(seedDomainMesh.vn == 0)
this->BuildMontecarloVolumeSampling(montecarloSampleNum);
if(this->seedDomainMesh.vn == 0)
tri::Append<MeshType,MeshType>::MeshCopy(seedDomainMesh,montecarloVolumeMesh);
vector<VertexPointer> pruningVec;
if(poissonRadius ==0 && poissonSampleNum!=0)
tri::PoissonPruningExact(seedDomainMesh,pruningVec,poissonRadius,poissonSampleNum,0.04,10,randSeed);
else
tri::PoissonPruning(seedDomainMesh,pruningVec,poissonRadius,randSeed);
std::vector<CoordType> seedPts(pruningVec.size());
for(size_t i=0;i<pruningVec.size();++i)
seedPts[i]=pruningVec[i]->P();
std::vector<CoordType> seedPts;
tri::PoissonPruning(seedDomainMesh,seedPts,poissonRadius,randSeed);
tri::BuildMeshFromCoordVector(this->seedMesh,seedPts);
// Kdtree must be rebuilt at the end of each step;
VertexConstDataWrapper<MeshType> vdw(seedMesh);
if(seedTree) delete seedTree;