trimesh_indexing.cpp

Go to the documentation of this file.

#include <iostream>
#include <QTime>
#include <omp.h>

#include "nanoflann.hpp"

#include <vcg/complex/complex.h>

#include <wrap/io_trimesh/import.h>
#include <wrap/io_trimesh/export.h>


#include <vcg/space/index/kdtree/kdtree.h>
#include <vcg/space/index/grid_static_ptr.h>
#include <vcg/space/index/perfect_spatial_hashing.h>
#include <vcg/space/index/spatial_hashing.h>
#include <vcg/space/index/octree.h>

int num_test = 1000;
int kNearest = 256;
float queryDist = 0.0037;
float ratio = 1000.0f;


class CVertex;
class CFace;
class CEdge;

class CUsedTypes        : public vcg::UsedTypes < vcg::Use< CVertex >::AsVertexType, vcg::Use< CFace >::AsFaceType>{};
class CVertex           : public vcg::Vertex < CUsedTypes, vcg::vertex::Coord3f, vcg::vertex::Normal3f, vcg::vertex::Radiusf, vcg::vertex::BitFlags, vcg::vertex::Qualityf, vcg::vertex::Color4b>{};
class CFace                     : public vcg::Face < CUsedTypes, vcg::face::VertexRef>{};

class CMesh                     : public vcg::tri::TriMesh < std::vector< CVertex >, std::vector< CFace > > {};


template <typename T>
struct PointCloud
{
        struct Point
        {
                T  x,y,z;
        };

        std::vector<Point>  pts;

        inline size_t kdtree_get_point_count() const { return pts.size(); }

        inline T kdtree_distance(const T *p1, const size_t idx_p2,size_t size) const
        {
                const T d0=p1[0]-pts[idx_p2].x;
                const T d1=p1[1]-pts[idx_p2].y;
                const T d2=p1[2]-pts[idx_p2].z;
                return d0*d0+d1*d1+d2*d2;
        }

        inline T kdtree_get_pt(const size_t idx, int dim) const
        {
                if (dim==0) return pts[idx].x;
                else if (dim==1) return pts[idx].y;
                else return pts[idx].z;
        }

        template <class BBOX>
        bool kdtree_get_bbox(BBOX &bb) const { return false; }

};


void testKDTree(CMesh& mesh, std::vector<unsigned int>& test_indeces, std::vector<vcg::Point3f>& randomSamples)
{
        std::cout << "==================================================="<< std::endl;
        std::cout << "KDTree" << std::endl;
        QTime time;
        time.start();

        // Construction of the kdTree
        vcg::ConstDataWrapper<CMesh::VertexType::CoordType> wrapperVcg(&mesh.vert[0].P(), mesh.vert.size(), size_t(mesh.vert[1].P().V()) - size_t(mesh.vert[0].P().V()));
        vcg::KdTree<CMesh::ScalarType> kdTreeVcg(wrapperVcg);
        std::cout << "Build: " << time.elapsed() <<  " ms" << std::endl;

        // Computation of the point radius
        float mAveragePointSpacing = 0;
        time.restart();
        #pragma omp parallel for reduction(+: mAveragePointSpacing) schedule(dynamic, 10)
        for (int i = 0; i < mesh.vert.size(); i++)
        {
                vcg::KdTree<CMesh::ScalarType>::PriorityQueue queue; 
                kdTreeVcg.doQueryK(mesh.vert[i].cP(), 16, queue);
                float newRadius = 2.0f * sqrt(queue.getWeight(0)/ queue.getNofElements());
                mesh.vert[i].R() -= newRadius;
                mAveragePointSpacing += newRadius;
        }
        mAveragePointSpacing /= mesh.vert.size();
        std::cout << "Average point radius (OpenMP) " << mAveragePointSpacing << std::endl;
        std::cout << "Time (OpenMP): " << time.elapsed() << " ms" << std::endl;
        
        queryDist = mAveragePointSpacing * 150;

        // Test with the radius search
        std::cout << "Radius search (" << num_test << " tests)"<< std::endl; 
        float avgTime = 0.0f;
        for (int ii = 0; ii < num_test; ii++)
        {
                time.restart();
                std::vector<unsigned int> indeces;
                std::vector<float> dists;
                kdTreeVcg.doQueryDist(mesh.vert[test_indeces[ii]].cP(), queryDist, indeces, dists);
                avgTime += time.elapsed();
        }
        std::cout << "Time (radius = " << queryDist << "): " << avgTime << " ms (mean " << avgTime / num_test << "ms)"  << std::endl;

        // Test with the k-nearest search
        std::cout << "k-Nearest search (" << num_test*10 << " tests)"<< std::endl;
        avgTime = 0.0f;
        for (int ii = 0; ii < num_test * 10; ii++)
        {
                time.restart();
                vcg::KdTree<CMesh::ScalarType>::PriorityQueue queue; 
                kdTreeVcg.doQueryK(mesh.vert[test_indeces[ii]].cP(), kNearest, queue);
                avgTime += time.elapsed();
        }
        std::cout << "Time (k = " << kNearest << "): " << avgTime << " ms (mean " << avgTime / (num_test * 10) << "ms)"  << std::endl;
        
        // Test with the closest search
        std::cout << "Closest search (" << num_test*10 << " tests)"<< std::endl;
        avgTime = 0.0f;
        for (int ii = 0; ii < num_test * 10; ii++)
        {
                time.restart();
                unsigned int index;
                float minDist;
                kdTreeVcg.doQueryClosest(randomSamples[ii], index, minDist);
                avgTime += time.elapsed();
        }
        std::cout << "Time : " << avgTime << " ms (mean " << avgTime / (num_test * 10) << "ms)"  << std::endl << std::endl;
}


void testNanoFLANN(CMesh& mesh, std::vector<unsigned int>& test_indeces, std::vector<vcg::Point3f> randomSamples)
{
        std::cout << "==================================================="<< std::endl;
        std::cout << "nanoFLANN" << std::endl;
        
        PointCloud<float> cloud;
        cloud.pts.resize(mesh.vert.size());
        for (size_t i=0; i < mesh.vert.size(); i++)
        {
                cloud.pts[i].x = mesh.vert[i].P().X();
                cloud.pts[i].y = mesh.vert[i].P().Y();
                cloud.pts[i].z = mesh.vert[i].P().Z();
        }

        typedef nanoflann::KDTreeSingleIndexAdaptor<
                nanoflann::L2_Simple_Adaptor<float, PointCloud<float> > ,
                PointCloud<float>,
                3 /* dim */
                > my_kd_tree_t;

        // Construction of the nanoFLANN KDtree
        QTime time;
        time.start();
        my_kd_tree_t   index(3, cloud, nanoflann::KDTreeSingleIndexAdaptorParams(16) );
        index.buildIndex();
        std::cout << "Build nanoFlann: " << time.elapsed() <<  " ms" << std::endl;

        // Test with the radius search
        std::cout << "Radius search (" << num_test << " tests)"<< std::endl; 
        float avgTime = 0.0f;
        std::vector<std::pair<size_t,float> >   ret_matches;
        nanoflann::SearchParams params;
        for (int ii = 0; ii < num_test; ii++)
        {
                time.restart();
                const size_t nMatches = index.radiusSearch(mesh.vert[test_indeces[ii]].P().V(), queryDist, ret_matches, params);
                avgTime += time.elapsed();
        }
        std::cout << "Time (radius = " << queryDist << "): " << avgTime << " ms (mean " << avgTime / num_test << "ms)"  << std::endl;

        // Test with the k-nearest search
        std::cout << "k-Nearest search (" << num_test*10 << " tests)"<< std::endl;
        avgTime = 0.0f;
        std::vector<size_t>   ret_index(kNearest);
        std::vector<float> out_dist_sqr(kNearest);
        for (int ii = 0; ii < num_test * 10; ii++)
        {
                time.restart();
                index.knnSearch(mesh.vert[test_indeces[ii]].P().V(), kNearest, &ret_index[0], &out_dist_sqr[0]);
                avgTime += time.elapsed();
        }
        std::cout << "Time (k = " << kNearest << "): " << avgTime << " ms (mean " << avgTime / (num_test * 10) << "ms)"  << std::endl;

        // Test with the closest search
        std::cout << "Closest search (" << num_test*10 << " tests)"<< std::endl;
        avgTime = 0.0f;
        std::vector<size_t>   ret_index_clos(1);
        std::vector<float> out_dist_sqr_clos(1);
        for (int ii = 0; ii < num_test * 10; ii++)
        {
                time.restart();
                index.knnSearch(randomSamples[ii].V(), 1, &ret_index_clos[0], &out_dist_sqr_clos[0]);
                avgTime += time.elapsed();
        }
        std::cout << "Time : " << avgTime << " ms (mean " << avgTime / (num_test * 10) << "ms)"  << std::endl << std::endl;
}


void testUniformGrid(CMesh& mesh, std::vector<unsigned int>& test_indeces, std::vector<vcg::Point3f>& randomSamples)
{
        std::cout << "==================================================="<< std::endl;
        std::cout << "Uniform Grid" << std::endl;
        QTime time;
        time.start();

        // Construction of the uniform grid
        typedef vcg::GridStaticPtr<CMesh::VertexType, CMesh::VertexType::ScalarType> MeshGrid; 
        MeshGrid uniformGrid;
        uniformGrid.Set(mesh.vert.begin(), mesh.vert.end());
        std::cout << "Build: " << time.elapsed() <<  " ms" << std::endl;

        // Test with the radius search
        std::cout << "Radius search (" << num_test << " tests)"<< std::endl; 
        float  avgTime = 0.0f;
        for (int ii = 0; ii < num_test; ii++)
        {
                time.restart();
                std::vector<CMesh::VertexPointer> vertexPtr;
                std::vector<CMesh::VertexType::CoordType> points;
                std::vector<float> dists;
                vcg::tri::GetInSphereVertex(mesh, uniformGrid, mesh.vert[test_indeces[ii]].cP(), queryDist, vertexPtr, dists, points);
                avgTime += time.elapsed();
        }
        std::cout << "Time (radius = " << queryDist << "): " << avgTime << " ms (mean " << avgTime / num_test << "ms)"  << std::endl;

        // Test with the k-nearest search
        std::cout << "k-Nearest search (" << num_test*10 << " tests)"<< std::endl;
        avgTime = 0.0f;
        for (int ii = 0; ii < num_test * 10; ii++)
        {
                time.restart();
                std::vector<CMesh::VertexPointer> vertexPtr;
                std::vector<CMesh::VertexType::CoordType> points;
                std::vector<float> dists;
                vcg::tri::GetKClosestVertex(mesh, uniformGrid, kNearest, mesh.vert[test_indeces[ii]].cP(), mesh.bbox.Diag(), vertexPtr, dists, points);
                avgTime += time.elapsed();
        }
        std::cout << "Time (k = " << kNearest << "): " << avgTime << " ms (mean " << avgTime / (num_test * 10) << "ms)"  << std::endl;

        // Test with the Closest search
        std::cout << "Closest search (" << num_test*10 << " tests)"<< std::endl;
        avgTime = 0.0f;
        for (int ii = 0; ii < num_test * 10; ii++)
        {
                time.restart();
                float minDist;
                vcg::tri::GetClosestVertex(mesh, uniformGrid, randomSamples[ii], mesh.bbox.Diag(), minDist);
                avgTime += time.elapsed();
        }
        std::cout << "Time : " << avgTime << " ms (mean " << avgTime / (num_test * 10) << "ms)"  << std::endl << std::endl;
}



void testSpatialHashing(CMesh& mesh, std::vector<unsigned int>& test_indeces, std::vector<vcg::Point3f>& randomSamples)
{
        std::cout << "==================================================="<< std::endl;
        std::cout << "Spatial Hashing" << std::endl;
        QTime time;
        time.start();

        // Construction of the uniform grid
        typedef vcg::SpatialHashTable<CMesh::VertexType, CMesh::VertexType::ScalarType> MeshGrid; 
        MeshGrid uniformGrid;
        uniformGrid.Set(mesh.vert.begin(), mesh.vert.end());
        std::cout << "Build: " << time.elapsed() <<  " ms" << std::endl;

        // Test with the radius search
        std::cout << "Radius search (" << num_test << " tests)"<< std::endl; 
        float  avgTime = 0.0f;
        for (int ii = 0; ii < num_test; ii++)
        {
                time.restart();
                std::vector<CMesh::VertexPointer> vertexPtr;
                std::vector<CMesh::VertexType::CoordType> points;
                std::vector<float> dists;
                vcg::tri::GetInSphereVertex(mesh, uniformGrid, mesh.vert[test_indeces[ii]].cP(), queryDist, vertexPtr, dists, points);
                avgTime += time.elapsed();
        }
        std::cout << "Time (radius = " << queryDist << "): " << avgTime << " ms (mean " << avgTime / num_test << "ms)"  << std::endl;

        // Test with the k-nearest search
        std::cout << "k-Nearest search (" << num_test*10 << " tests)"<< std::endl;
        avgTime = 0.0f;
        for (int ii = 0; ii < num_test * 10; ii++)
        {
                time.restart();
                std::vector<CMesh::VertexPointer> vertexPtr;
                std::vector<CMesh::VertexType::CoordType> points;
                std::vector<float> dists;
                vcg::tri::GetKClosestVertex(mesh, uniformGrid, kNearest, mesh.vert[test_indeces[ii]].cP(), mesh.bbox.Diag(), vertexPtr, dists, points);
                avgTime += time.elapsed();
        }
        std::cout << "Time (k = " << kNearest << "): " << avgTime << " ms (mean " << avgTime / (num_test * 10) << "ms)"  << std::endl;

        // Test with the Closest search
        std::cout << "Closest search (" << num_test*10 << " tests)"<< std::endl;
        avgTime = 0.0f;
        for (int ii = 0; ii < num_test * 10; ii++)
        {
                time.restart();
                float minDist;
                vcg::tri::GetClosestVertex(mesh, uniformGrid, randomSamples[ii], mesh.bbox.Diag(), minDist);
                avgTime += time.elapsed();
        }
        std::cout << "Time : " << avgTime << " ms (mean " << avgTime / (num_test * 10) << "ms)"  << std::endl << std::endl;
}



void testPerfectSpatialHashing(CMesh& mesh, std::vector<unsigned int>& test_indeces)
{
        std::cout << "==================================================="<< std::endl;
        std::cout << "Perfect Spatial Hashing" << std::endl;
        QTime time;
        time.start();

        // Construction of the uniform grid
        typedef vcg::SpatialHashTable<CMesh::VertexType, CMesh::VertexType::ScalarType> MeshGrid; 
        MeshGrid uniformGrid;
        uniformGrid.Set(mesh.vert.begin(), mesh.vert.end());
        std::cout << "Build: " << time.elapsed() <<  " ms" << std::endl;

        // Test with the radius search
        std::cout << "Radius search (" << num_test << " tests)"<< std::endl; 
        float  avgTime = 0.0f;
        for (int ii = 0; ii < num_test; ii++)
        {
                time.restart();
                std::vector<CMesh::VertexPointer> vertexPtr;
                std::vector<CMesh::VertexType::CoordType> points;
                std::vector<float> dists;
                vcg::tri::GetInSphereVertex(mesh, uniformGrid, mesh.vert[test_indeces[ii]].cP(), queryDist, vertexPtr, dists, points);
                avgTime += time.elapsed();
        }
        std::cout << "Time (radius = " << queryDist << "): " << avgTime << " ms (mean " << avgTime / num_test << "ms)"  << std::endl << std::endl;
}


void testOctree(CMesh& mesh, std::vector<unsigned int>& test_indeces, std::vector<vcg::Point3f>& randomSamples)
{
        std::cout << "==================================================="<< std::endl;
        std::cout << "Octree" << std::endl;
        QTime time;
        time.start();

        // Construction of the uniform grid
        typedef vcg::Octree<CMesh::VertexType, CMesh::VertexType::ScalarType> MeshGrid; 
        MeshGrid uniformGrid;
        uniformGrid.Set(mesh.vert.begin(), mesh.vert.end());
        std::cout << "Build: " << time.elapsed() <<  " ms" << std::endl;

        // Test with the radius search
        std::cout << "Radius search (" << num_test << " tests)"<< std::endl; 
        float  avgTime = 0.0f;
        for (int ii = 0; ii < num_test; ii++)
        {
                time.restart();
                std::vector<CMesh::VertexPointer> vertexPtr;
                std::vector<CMesh::VertexType::CoordType> points;
                std::vector<float> dists;
                vcg::tri::GetInSphereVertex(mesh, uniformGrid, mesh.vert[test_indeces[ii]].cP(), queryDist, vertexPtr, dists, points);
                avgTime += time.elapsed();
        }
        std::cout << "Time (radius = " << queryDist << "): " << avgTime << " ms (mean " << avgTime / num_test << "ms)"  << std::endl;

        // Test with the k-nearest search
        std::cout << "k-Nearest search (" << num_test*10 << " tests)"<< std::endl;
        avgTime = 0.0f;
        for (int ii = 0; ii < num_test * 10; ii++)
        {
                time.restart();
                std::vector<CMesh::VertexPointer> vertexPtr;
                std::vector<CMesh::VertexType::CoordType> points;
                std::vector<float> dists;
                vcg::tri::GetKClosestVertex(mesh, uniformGrid, kNearest, mesh.vert[test_indeces[ii]].cP(), mesh.bbox.Diag(), vertexPtr, dists, points);
                avgTime += time.elapsed();
        }
        std::cout << "Time (k = " << kNearest << "): " << avgTime << " ms (mean " << avgTime / (num_test * 10) << "ms)"  << std::endl;

        // Test with the Closest search
        std::cout << "Closest search (" << num_test*10 << " tests)"<< std::endl;
        avgTime = 0.0f;
        for (int ii = 0; ii < num_test * 10; ii++)
        {
                time.restart();
                float minDist;
                vcg::tri::GetClosestVertex(mesh, uniformGrid, randomSamples[ii], mesh.bbox.Diag(), minDist);
                avgTime += time.elapsed();
        }
        std::cout << "Time : " << avgTime << " ms (mean " << avgTime / (num_test * 10) << "ms)"  << std::endl << std::endl;
}



int main( int argc, char * argv[] )
{
        if (argc < 2)
                std::cout << "Invalid arguments" << std::endl;

        CMesh mesh;
        if (vcg::tri::io::Importer<CMesh>::Open(mesh, argv[1])  != 0)
                std::cout << "Invalid filename" << std::endl;

        std::cout << "Mesh BBox diagonal: " << mesh.bbox.Diag() << std::endl; 
        std::cout << "Max point random offset: " << mesh.bbox.Diag() / 1000.0f << std::endl << std::endl; 

        vcg::math::MarsenneTwisterRNG randGen;
        randGen.initialize(0);
        std::vector<vcg::Point3f> randomSamples;
        for (int i = 0; i < num_test * 10; i++)
                randomSamples.push_back(vcg::math::GeneratePointOnUnitSphereUniform<float>(randGen) * randGen.generate01() * mesh.bbox.Diag() / ratio);

        std::vector<unsigned int> test_indeces;
        for (int i = 0; i < num_test * 10; i++)
        {
                int index = randGen.generate01() * (mesh.vert.size() - 1);
                test_indeces.push_back(index);
                randomSamples[i] += mesh.vert[i].P();
        }

        testKDTree(mesh, test_indeces, randomSamples);
        testNanoFLANN(mesh, test_indeces, randomSamples);
        testUniformGrid(mesh, test_indeces, randomSamples);
        testSpatialHashing(mesh, test_indeces, randomSamples);
        testPerfectSpatialHashing(mesh, test_indeces);
        testOctree(mesh, test_indeces, randomSamples);
}