SpectralDecomposition.cpp

Go to the documentation of this file.

// kate: replace-tabs off; indent-width 4; indent-mode normal
// vim: ts=4:sw=4:noexpandtab
/*

Copyright (c) 2010--2018,
François Pomerleau and Stephane Magnenat, ASL, ETHZ, Switzerland
You can contact the authors at <f dot pomerleau at gmail dot com> and
<stephane at magnenat dot net>

All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
    * Redistributions of source code must retain the above copyright
      notice, this list of conditions and the following disclaimer.
    * Redistributions in binary form must reproduce the above copyright
      notice, this list of conditions and the following disclaimer in the
      documentation and/or other materials provided with the distribution.
    * Neither the name of the <organization> nor the
      names of its contributors may be used to endorse or promote products
      derived from this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL ETH-ASL BE LIABLE FOR ANY
DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

*/
#include "SpectralDecomposition.h"

#include <random>

// SpectralDecomposition
template <typename T>
SpectralDecompositionDataPointsFilter<T>::SpectralDecompositionDataPointsFilter(const Parameters& params) :
        PointMatcher<T>::DataPointsFilter("SpectralDecompositionDataPointsFilter", 
                SpectralDecompositionDataPointsFilter::availableParameters(), params),
        k{Parametrizable::get<std::size_t>("k")},
        sigma{Parametrizable::get<T>("sigma")},
        radius{Parametrizable::get<T>("radius")},
        itMax{Parametrizable::get<std::size_t>("itMax")},
        keepNormals{Parametrizable::get<bool>("keepNormals")},
        keepLabels{Parametrizable::get<bool>("keepLabels")},
        keepLambdas{Parametrizable::get<bool>("keepLambdas")},
        keepTensors{Parametrizable::get<bool>("keepTensors")}
{
}

template <typename T>
typename PointMatcher<T>::DataPoints
SpectralDecompositionDataPointsFilter<T>::filter(const DataPoints& input)
{
        DataPoints output(input);
        inPlaceFilter(output);
        return output;
}

template <typename T>
void SpectralDecompositionDataPointsFilter<T>::inPlaceFilter(DataPoints& cloud)
{
        const std::size_t nbPts = cloud.getNbPoints();
        
        if(k > nbPts) return;
        
        TensorVoting<T> tv{sigma, k};
        
//--- 1. Vote to determine prefered orientation + density estimation -----------
        tv.encode(cloud, TensorVoting<T>::Encoding::BALL);
        tv.cfvote(cloud, true);
        tv.decompose();
        tv.toDescriptors();
        addDescriptor(cloud, tv, false /*normals*/, false /*labels*/, true /*lambdas*/, false /*tensors*/);
        
//--- 2. Filter iteratively on each measure (surfaceness, curveness, pointness) to uniformize density
        std::size_t it = 0;
        const std::size_t itMax_ = itMax;
        const std::size_t k_ = k;
        std::size_t oldnbPts = nbPts;
        
        auto checkConvergence = [&oldnbPts, &it, &itMax_, &k_](const DataPoints& pts, const std::size_t threshold) mutable ->bool{              
                const std::size_t nbPts = pts.getNbPoints();
                bool ret = (oldnbPts - nbPts) < threshold;
                
                oldnbPts = nbPts;
                
                return ret or ++it >= itMax_ or k_ >= nbPts;            
        };
        
        const T xi3 = xi_expectation(3, sigma, radius);
        const T xi2 = xi_expectation(2, sigma, radius);
        const T xi1 = xi_expectation(1, sigma, radius);
        
        do 
        {       
        // 2.1 On pointness
                filterPointness(cloud, xi3, tv.k);
        // 2.2 On curveness
                filterCurveness(cloud, xi1, tv.k);
        // 2.3 On surfaceness
                filterSurfaceness(cloud, xi2, tv.k);

        //Re-compute vote...
                tv.encode(cloud, TensorVoting<T>::Encoding::BALL);
                tv.cfvote(cloud, true);
                tv.decompose();
                tv.toDescriptors();
                
                addDescriptor(cloud, tv, false /*normals*/, false /*labels*/, true /*lambdas*/, false /*tensors*/);
        } 
        while(not checkConvergence(cloud, 5 /*delta points*/));
        
//--- 3. Re-encode as Aware tensors + Re-vote ----------------------------------
        addDescriptor(cloud, tv, false /*normals*/, false /*labels*/, false /*lambdas*/, true /*tensors*/);
        tv.encode(cloud, TensorVoting<T>::Encoding::AWARE_TENSOR);
        tv.cfvote(cloud);
        tv.decompose();
        tv.toDescriptors();

//--- 4. Add descriptors
        addDescriptor(cloud, tv, keepNormals, true, keepLambdas, keepTensors);//TODO: add remove not kept descriptors
        
//--- 5. Remove outliers
        removeOutlier(cloud, tv);
}


template <typename T>
void SpectralDecompositionDataPointsFilter<T>::addDescriptor(DataPoints& pts, const TensorVoting<T> &tv, bool keepNormals_, bool keepLabels_, bool keepLambdas_, bool keepTensors_) const
{
        const std::size_t nbPts = pts.getNbPoints();

        Matrix labels  = Matrix::Zero(1, nbPts);
        Matrix l1  = PM::Matrix::Zero(1, nbPts);        
        Matrix l2  = PM::Matrix::Zero(1, nbPts);        
        Matrix l3  = PM::Matrix::Zero(1, nbPts);
        
        if(keepLabels_ or keepLambdas_)
        {
        #pragma omp parallel for
                for(std::size_t i = 0; i < nbPts; ++i)
                {
                        const T lambda1 = tv.surfaceness(i) + tv.curveness(i) + tv.pointness(i);
                        const T lambda2 = tv.curveness(i) + tv.pointness(i);
                        const T lambda3 = tv.pointness(i);
        
                        int index;
                        Vector coeff = (Vector(3) << lambda3, (lambda2 - lambda3), (lambda1 - lambda2)).finished();
                        coeff.maxCoeff(&index);  

                        labels(i) = index + 1 ;
                
                        l1(i) = lambda1 * k;
                        l2(i) = lambda2 * k;
                        l3(i) = lambda3 * k;
                }
        }
        try
        {       
            pts.addDescriptor("surfaceness", tv.surfaceness);
            pts.addDescriptor("curveness", tv.curveness);
            pts.addDescriptor("pointness", tv.pointness);
            
            if(keepLambdas_)
            {
                        pts.addDescriptor("lambda1", l1);
                        pts.addDescriptor("lambda2", l2);
                        pts.addDescriptor("lambda3", l3);
            }
            
                if(keepNormals_)
                {    
                        pts.addDescriptor("normals", tv.normals);
                        pts.addDescriptor("tangents", tv.tangents);
                }    
            if(keepLabels_)
            {
                        pts.addDescriptor("labels", labels);
            }  
            if(keepTensors_)
            {
                        pts.addDescriptor("sticks", tv.sticks);
                        pts.addDescriptor("plates", tv.plates);
                        pts.addDescriptor("balls", tv.balls);
            }
        }
        catch (...) {
            std::cerr << "SpectralDecomposition<T>::inPlaceFilter::addDescriptor: Cannot add descriptors to pointcloud" << std::endl;
        }

}


//------------------------------------------------------------------------------
// Outlier filter
//------------------------------------------------------------------------------
template <typename T>
void SpectralDecompositionDataPointsFilter<T>::removeOutlier(DataPoints& pts, const TensorVoting<T> &tv) const
{
        static constexpr int POINT = 0;
        static constexpr int CURVE = 1;
        static constexpr int SURFACE = 2;
        
        static constexpr T th = 0.1; //threshold at 10%
        
        const std::size_t nbPts = pts.getNbPoints();
                
        const T th_p = (tv.pointness.maxCoeff() - tv.pointness.minCoeff()) * th + tv.pointness.minCoeff();
        const T th_c = (tv.curveness.maxCoeff() - tv.curveness.minCoeff()) * th + tv.curveness.minCoeff();
        const T th_s = (tv.surfaceness.maxCoeff() - tv.surfaceness.minCoeff()) * th + tv.surfaceness.minCoeff();
        
        
        std::size_t j = 0;
        for (std::size_t i = 0; i < nbPts; ++i)
        {
                const T surfaceness = tv.surfaceness(i);
                const T curveness = tv.curveness(i);
                const T pointness = tv.pointness(i);

                int label;
                (Vector(3) << pointness, curveness, surfaceness).finished().maxCoeff(&label); 

                bool keepPt = ((label == POINT) and  (pointness >= th_p)) 
                        or ((label == CURVE) and  (curveness >= th_c)) 
                        or ((label == SURFACE) and  (surfaceness >= th_s));
                
                if (keepPt)
                {
                        pts.setColFrom(j, pts, i);
                        ++j;
                }
        }
        pts.conservativeResize(j);      
}

//------------------------------------------------------------------------------
// Filter
//------------------------------------------------------------------------------
template <typename T>
void SpectralDecompositionDataPointsFilter<T>::filterSurfaceness(DataPoints& pts, T xi, std::size_t k) const
{
        constexpr std::size_t seed = 1;
        std::mt19937 gen(seed); //Standard mersenne_twister_engine seeded with seed
        std::uniform_real_distribution<> uni01(0., 1.);
        
        const std::size_t nbPts = pts.getNbPoints();
        
        // Check field exists
        if (!pts.descriptorExists("lambda1") or !pts.descriptorExists("lambda2") or !pts.descriptorExists("lambda3"))
        {
                throw InvalidField("SpectralDecomposition<T>::filter: Error, lambdas field not found in descriptors.");
        }
        
        const auto& lambda1 = pts.getDescriptorViewByName("lambda1");
        const auto& lambda2 = pts.getDescriptorViewByName("lambda2");
        const auto& lambda3 = pts.getDescriptorViewByName("lambda3");

        std::size_t j = 0;
        for (std::size_t i = 0; i < nbPts; ++i)
        {
                const T randv = uni01(gen);     
                
                const T nl1 = lambda1(0,i) / k;
                const T nl2 = lambda2(0,i) / k;
                const T nl3 = lambda3(0,i) / k;
                
                if (nl1 < xi or nl2 < 0.75 * xi or nl3 < 0.75 * xi or randv < 0.5)
                {
                        pts.setColFrom(j, pts, i);
                        ++j;
                }
        }
        pts.conservativeResize(j);
}

template <typename T>
void SpectralDecompositionDataPointsFilter<T>::filterCurveness(DataPoints& pts, T xi, std::size_t k) const
{
        constexpr std::size_t seed = 1;
        std::mt19937 gen(seed); //Standard mersenne_twister_engine seeded with seed
        std::uniform_real_distribution<> uni01(0., 1.);
        
        const std::size_t nbPts = pts.getNbPoints();
        
        // Check field exists
        if (!pts.descriptorExists("lambda1") or !pts.descriptorExists("lambda2") or !pts.descriptorExists("lambda3"))
        {
                throw InvalidField("SpectralDecomposition<T>::filter: Error, lambdas field not found in descriptors.");
        }
        
        const auto& lambda1 = pts.getDescriptorViewByName("lambda1");
        const auto& lambda2 = pts.getDescriptorViewByName("lambda2");
        const auto& lambda3 = pts.getDescriptorViewByName("lambda3");

        std::size_t j = 0;
        for (std::size_t i = 0; i < nbPts; ++i)
        {
                const T randv = uni01(gen);     
                
                const T nl1 = lambda1(0,i) / k;
                const T nl2 = lambda2(0,i) / k;
                const T nl3 = lambda3(0,i) / k;
                
                if (nl1 < xi or nl2 < xi or nl3 < 0.5 * xi or randv < 0.5)
                {
                        pts.setColFrom(j, pts, i);
                        ++j;
                }
        }
        pts.conservativeResize(j);
}

template <typename T>
void SpectralDecompositionDataPointsFilter<T>::filterPointness(DataPoints& pts, T xi, std::size_t k) const
{
        constexpr std::size_t seed = 1;
        std::mt19937 gen(seed); //Standard mersenne_twister_engine seeded with seed
        std::uniform_real_distribution<> uni01(0., 1.);
        
        const std::size_t nbPts = pts.getNbPoints();
        
        // Check field exists
        if (!pts.descriptorExists("lambda1") or !pts.descriptorExists("lambda2") or !pts.descriptorExists("lambda3"))
        {
                throw InvalidField("SpectralDecomposition<T>::filter: Error, lambdas field not found in descriptors.");
        }
        
        const auto& lambda1 = pts.getDescriptorViewByName("lambda1");
        const auto& lambda2 = pts.getDescriptorViewByName("lambda2");
        const auto& lambda3 = pts.getDescriptorViewByName("lambda3");

        std::size_t j = 0;
        for (std::size_t i = 0; i < nbPts; ++i)
        {
                const T randv = uni01(gen);     
                
                const T nl1 = lambda1(0,i) / k;
                const T nl2 = lambda2(0,i) / k;
                const T nl3 = lambda3(0,i) / k;
                
                if (nl1 < (5./6.) * xi or nl2 < (5./6.) * xi or nl3 < (5./6.) * xi or randv < 0.2)
                {
                        pts.setColFrom(j, pts, i);
                        ++j;
                }
        }
        pts.conservativeResize(j);
}

template struct SpectralDecompositionDataPointsFilter<float>;
template struct SpectralDecompositionDataPointsFilter<double>;