Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance.
Project price only 1 $
You can buy this project and download/modify it how often you want.
srcnativelibs.Include.OpenCV.opencv2.flann.kmeans_index.h Maven / Gradle / Ivy
/***********************************************************************
* Software License Agreement (BSD License)
*
* Copyright 2008-2009 Marius Muja ([email protected] ). All rights reserved.
* Copyright 2008-2009 David G. Lowe ([email protected] ). All rights reserved.
*
* THE BSD LICENSE
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. 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.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 THE AUTHOR 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.
*************************************************************************/
#ifndef OPENCV_FLANN_KMEANS_INDEX_H_
#define OPENCV_FLANN_KMEANS_INDEX_H_
#include
#include
#include
#include
#include
#include
#include "general.h"
#include "nn_index.h"
#include "dist.h"
#include "matrix.h"
#include "result_set.h"
#include "heap.h"
#include "allocator.h"
#include "random.h"
#include "saving.h"
#include "logger.h"
namespace cvflann
{
struct KMeansIndexParams : public IndexParams
{
KMeansIndexParams(int branching = 32, int iterations = 11,
flann_centers_init_t centers_init = FLANN_CENTERS_RANDOM, float cb_index = 0.2 )
{
(*this)["algorithm"] = FLANN_INDEX_KMEANS;
// branching factor
(*this)["branching"] = branching;
// max iterations to perform in one kmeans clustering (kmeans tree)
(*this)["iterations"] = iterations;
// algorithm used for picking the initial cluster centers for kmeans tree
(*this)["centers_init"] = centers_init;
// cluster boundary index. Used when searching the kmeans tree
(*this)["cb_index"] = cb_index;
}
};
/**
* Hierarchical kmeans index
*
* Contains a tree constructed through a hierarchical kmeans clustering
* and other information for indexing a set of points for nearest-neighbour matching.
*/
template
class KMeansIndex : public NNIndex
{
public:
typedef typename Distance::ElementType ElementType;
typedef typename Distance::ResultType DistanceType;
typedef void (KMeansIndex::* centersAlgFunction)(int, int*, int, int*, int&);
/**
* The function used for choosing the cluster centers.
*/
centersAlgFunction chooseCenters;
/**
* Chooses the initial centers in the k-means clustering in a random manner.
*
* Params:
* k = number of centers
* vecs = the dataset of points
* indices = indices in the dataset
* indices_length = length of indices vector
*
*/
void chooseCentersRandom(int k, int* indices, int indices_length, int* centers, int& centers_length)
{
UniqueRandom r(indices_length);
int index;
for (index=0; index=0 && rnd < n);
centers[0] = indices[rnd];
int index;
for (index=1; indexbest_val) {
best_val = dist;
best_index = j;
}
}
if (best_index!=-1) {
centers[index] = indices[best_index];
}
else {
break;
}
}
centers_length = index;
}
/**
* Chooses the initial centers in the k-means using the algorithm
* proposed in the KMeans++ paper:
* Arthur, David; Vassilvitskii, Sergei - k-means++: The Advantages of Careful Seeding
*
* Implementation of this function was converted from the one provided in Arthur's code.
*
* Params:
* k = number of centers
* vecs = the dataset of points
* indices = indices in the dataset
* Returns:
*/
void chooseCentersKMeanspp(int k, int* indices, int indices_length, int* centers, int& centers_length)
{
int n = indices_length;
double currentPot = 0;
DistanceType* closestDistSq = new DistanceType[n];
// Choose one random center and set the closestDistSq values
int index = rand_int(n);
assert(index >=0 && index < n);
centers[0] = indices[index];
for (int i = 0; i < n; i++) {
closestDistSq[i] = distance_(dataset_[indices[i]], dataset_[indices[index]], dataset_.cols);
currentPot += closestDistSq[i];
}
const int numLocalTries = 1;
// Choose each center
int centerCount;
for (centerCount = 1; centerCount < k; centerCount++) {
// Repeat several trials
double bestNewPot = -1;
int bestNewIndex = -1;
for (int localTrial = 0; localTrial < numLocalTries; localTrial++) {
// Choose our center - have to be slightly careful to return a valid answer even accounting
// for possible rounding errors
double randVal = rand_double(currentPot);
for (index = 0; index < n-1; index++) {
if (randVal <= closestDistSq[index]) break;
else randVal -= closestDistSq[index];
}
// Compute the new potential
double newPot = 0;
for (int i = 0; i < n; i++) newPot += std::min( distance_(dataset_[indices[i]], dataset_[indices[index]], dataset_.cols), closestDistSq[i] );
// Store the best result
if ((bestNewPot < 0)||(newPot < bestNewPot)) {
bestNewPot = newPot;
bestNewIndex = index;
}
}
// Add the appropriate center
centers[centerCount] = indices[bestNewIndex];
currentPot = bestNewPot;
for (int i = 0; i < n; i++) closestDistSq[i] = std::min( distance_(dataset_[indices[i]], dataset_[indices[bestNewIndex]], dataset_.cols), closestDistSq[i] );
}
centers_length = centerCount;
delete[] closestDistSq;
}
public:
flann_algorithm_t getType() const
{
return FLANN_INDEX_KMEANS;
}
/**
* Index constructor
*
* Params:
* inputData = dataset with the input features
* params = parameters passed to the hierarchical k-means algorithm
*/
KMeansIndex(const Matrix& inputData, const IndexParams& params = KMeansIndexParams(),
Distance d = Distance())
: dataset_(inputData), index_params_(params), root_(NULL), indices_(NULL), distance_(d)
{
memoryCounter_ = 0;
size_ = dataset_.rows;
veclen_ = dataset_.cols;
branching_ = get_param(params,"branching",32);
iterations_ = get_param(params,"iterations",11);
if (iterations_<0) {
iterations_ = (std::numeric_limits::max)();
}
centers_init_ = get_param(params,"centers_init",FLANN_CENTERS_RANDOM);
if (centers_init_==FLANN_CENTERS_RANDOM) {
chooseCenters = &KMeansIndex::chooseCentersRandom;
}
else if (centers_init_==FLANN_CENTERS_GONZALES) {
chooseCenters = &KMeansIndex::chooseCentersGonzales;
}
else if (centers_init_==FLANN_CENTERS_KMEANSPP) {
chooseCenters = &KMeansIndex::chooseCentersKMeanspp;
}
else {
throw FLANNException("Unknown algorithm for choosing initial centers.");
}
cb_index_ = 0.4f;
}
KMeansIndex(const KMeansIndex&);
KMeansIndex& operator=(const KMeansIndex&);
/**
* Index destructor.
*
* Release the memory used by the index.
*/
virtual ~KMeansIndex()
{
if (root_ != NULL) {
free_centers(root_);
}
if (indices_!=NULL) {
delete[] indices_;
}
}
/**
* Returns size of index.
*/
size_t size() const
{
return size_;
}
/**
* Returns the length of an index feature.
*/
size_t veclen() const
{
return veclen_;
}
void set_cb_index( float index)
{
cb_index_ = index;
}
/**
* Computes the inde memory usage
* Returns: memory used by the index
*/
int usedMemory() const
{
return pool_.usedMemory+pool_.wastedMemory+memoryCounter_;
}
/**
* Builds the index
*/
void buildIndex()
{
if (branching_<2) {
throw FLANNException("Branching factor must be at least 2");
}
indices_ = new int[size_];
for (size_t i=0; i();
computeNodeStatistics(root_, indices_, (int)size_);
computeClustering(root_, indices_, (int)size_, branching_,0);
}
void saveIndex(FILE* stream)
{
save_value(stream, branching_);
save_value(stream, iterations_);
save_value(stream, memoryCounter_);
save_value(stream, cb_index_);
save_value(stream, *indices_, (int)size_);
save_tree(stream, root_);
}
void loadIndex(FILE* stream)
{
load_value(stream, branching_);
load_value(stream, iterations_);
load_value(stream, memoryCounter_);
load_value(stream, cb_index_);
if (indices_!=NULL) {
delete[] indices_;
}
indices_ = new int[size_];
load_value(stream, *indices_, size_);
if (root_!=NULL) {
free_centers(root_);
}
load_tree(stream, root_);
index_params_["algorithm"] = getType();
index_params_["branching"] = branching_;
index_params_["iterations"] = iterations_;
index_params_["centers_init"] = centers_init_;
index_params_["cb_index"] = cb_index_;
}
/**
* Find set of nearest neighbors to vec. Their indices are stored inside
* the result object.
*
* Params:
* result = the result object in which the indices of the nearest-neighbors are stored
* vec = the vector for which to search the nearest neighbors
* searchParams = parameters that influence the search algorithm (checks, cb_index)
*/
void findNeighbors(ResultSet& result, const ElementType* vec, const SearchParams& searchParams)
{
int maxChecks = get_param(searchParams,"checks",32);
if (maxChecks==FLANN_CHECKS_UNLIMITED) {
findExactNN(root_, result, vec);
}
else {
// Priority queue storing intermediate branches in the best-bin-first search
Heap* heap = new Heap((int)size_);
int checks = 0;
findNN(root_, result, vec, checks, maxChecks, heap);
BranchSt branch;
while (heap->popMin(branch) && (checks& centers)
{
int numClusters = centers.rows;
if (numClusters<1) {
throw FLANNException("Number of clusters must be at least 1");
}
DistanceType variance;
KMeansNodePtr* clusters = new KMeansNodePtr[numClusters];
int clusterCount = getMinVarianceClusters(root_, clusters, numClusters, variance);
Logger::info("Clusters requested: %d, returning %d\n",numClusters, clusterCount);
for (int i=0; ipivot;
for (size_t j=0; j BranchSt;
void save_tree(FILE* stream, KMeansNodePtr node)
{
save_value(stream, *node);
save_value(stream, *(node->pivot), (int)veclen_);
if (node->childs==NULL) {
int indices_offset = (int)(node->indices - indices_);
save_value(stream, indices_offset);
}
else {
for(int i=0; ichilds[i]);
}
}
}
void load_tree(FILE* stream, KMeansNodePtr& node)
{
node = pool_.allocate();
load_value(stream, *node);
node->pivot = new DistanceType[veclen_];
load_value(stream, *(node->pivot), (int)veclen_);
if (node->childs==NULL) {
int indices_offset;
load_value(stream, indices_offset);
node->indices = indices_ + indices_offset;
}
else {
node->childs = pool_.allocate(branching_);
for(int i=0; ichilds[i]);
}
}
}
/**
* Helper function
*/
void free_centers(KMeansNodePtr node)
{
delete[] node->pivot;
if (node->childs!=NULL) {
for (int k=0; kchilds[k]);
}
}
}
/**
* Computes the statistics of a node (mean, radius, variance).
*
* Params:
* node = the node to use
* indices = the indices of the points belonging to the node
*/
void computeNodeStatistics(KMeansNodePtr node, int* indices, int indices_length)
{
DistanceType radius = 0;
DistanceType variance = 0;
DistanceType* mean = new DistanceType[veclen_];
memoryCounter_ += int(veclen_*sizeof(DistanceType));
memset(mean,0,veclen_*sizeof(DistanceType));
for (size_t i=0; i(), veclen_);
}
for (size_t j=0; j(), veclen_);
DistanceType tmp = 0;
for (int i=0; iradius) {
radius = tmp;
}
}
node->variance = variance;
node->radius = radius;
node->pivot = mean;
}
/**
* The method responsible with actually doing the recursive hierarchical
* clustering
*
* Params:
* node = the node to cluster
* indices = indices of the points belonging to the current node
* branching = the branching factor to use in the clustering
*
* TODO: for 1-sized clusters don't store a cluster center (it's the same as the single cluster point)
*/
void computeClustering(KMeansNodePtr node, int* indices, int indices_length, int branching, int level)
{
node->size = indices_length;
node->level = level;
if (indices_length < branching) {
node->indices = indices;
std::sort(node->indices,node->indices+indices_length);
node->childs = NULL;
return;
}
int* centers_idx = new int[branching];
int centers_length;
(this->*chooseCenters)(branching, indices, indices_length, centers_idx, centers_length);
if (centers_lengthindices = indices;
std::sort(node->indices,node->indices+indices_length);
node->childs = NULL;
delete [] centers_idx;
return;
}
Matrix dcenters(new double[branching*veclen_],branching,veclen_);
for (int i=0; i radiuses(branching);
int* count = new int[branching];
for (int i=0; inew_sq_dist) {
belongs_to[i] = j;
sq_dist = new_sq_dist;
}
}
if (sq_dist>radiuses[belongs_to[i]]) {
radiuses[belongs_to[i]] = sq_dist;
}
count[belongs_to[i]]++;
}
bool converged = false;
int iteration = 0;
while (!converged && iterationnew_sq_dist) {
new_centroid = j;
sq_dist = new_sq_dist;
}
}
if (sq_dist>radiuses[new_centroid]) {
radiuses[new_centroid] = sq_dist;
}
if (new_centroid != belongs_to[i]) {
count[belongs_to[i]]--;
count[new_centroid]++;
belongs_to[i] = new_centroid;
converged = false;
}
}
for (int i=0; ichilds = pool_.allocate(branching);
int start = 0;
int end = start;
for (int c=0; c(), veclen_);
variance += d;
mean_radius += sqrt(d);
std::swap(indices[i],indices[end]);
std::swap(belongs_to[i],belongs_to[end]);
end++;
}
}
variance /= s;
mean_radius /= s;
variance -= distance_(centers[c], ZeroIterator(), veclen_);
node->childs[c] = pool_.allocate();
node->childs[c]->radius = radiuses[c];
node->childs[c]->pivot = centers[c];
node->childs[c]->variance = variance;
node->childs[c]->mean_radius = mean_radius;
node->childs[c]->indices = NULL;
computeClustering(node->childs[c],indices+start, end-start, branching, level+1);
start=end;
}
delete[] dcenters.data;
delete[] centers;
delete[] count;
delete[] belongs_to;
}
/**
* Performs one descent in the hierarchical k-means tree. The branches not
* visited are stored in a priority queue.
*
* Params:
* node = node to explore
* result = container for the k-nearest neighbors found
* vec = query points
* checks = how many points in the dataset have been checked so far
* maxChecks = maximum dataset points to checks
*/
void findNN(KMeansNodePtr node, ResultSet& result, const ElementType* vec, int& checks, int maxChecks,
Heap* heap)
{
// Ignore those clusters that are too far away
{
DistanceType bsq = distance_(vec, node->pivot, veclen_);
DistanceType rsq = node->radius;
DistanceType wsq = result.worstDist();
DistanceType val = bsq-rsq-wsq;
DistanceType val2 = val*val-4*rsq*wsq;
//if (val>0) {
if ((val>0)&&(val2>0)) {
return;
}
}
if (node->childs==NULL) {
if (checks>=maxChecks) {
if (result.full()) return;
}
checks += node->size;
for (int i=0; isize; ++i) {
int index = node->indices[i];
DistanceType dist = distance_(dataset_[index], vec, veclen_);
result.addPoint(dist, index);
}
}
else {
DistanceType* domain_distances = new DistanceType[branching_];
int closest_center = exploreNodeBranches(node, vec, domain_distances, heap);
delete[] domain_distances;
findNN(node->childs[closest_center],result,vec, checks, maxChecks, heap);
}
}
/**
* Helper function that computes the nearest childs of a node to a given query point.
* Params:
* node = the node
* q = the query point
* distances = array with the distances to each child node.
* Returns:
*/
int exploreNodeBranches(KMeansNodePtr node, const ElementType* q, DistanceType* domain_distances, Heap* heap)
{
int best_index = 0;
domain_distances[best_index] = distance_(q, node->childs[best_index]->pivot, veclen_);
for (int i=1; ichilds[i]->pivot, veclen_);
if (domain_distances[i]childs[best_index]->pivot;
for (int i=0; ichilds[i]->variance;
// float dist_to_border = getDistanceToBorder(node.childs[i].pivot,best_center,q);
// if (domain_distances[i]insert(BranchSt(node->childs[i],domain_distances[i]));
}
}
return best_index;
}
/**
* Function the performs exact nearest neighbor search by traversing the entire tree.
*/
void findExactNN(KMeansNodePtr node, ResultSet& result, const ElementType* vec)
{
// Ignore those clusters that are too far away
{
DistanceType bsq = distance_(vec, node->pivot, veclen_);
DistanceType rsq = node->radius;
DistanceType wsq = result.worstDist();
DistanceType val = bsq-rsq-wsq;
DistanceType val2 = val*val-4*rsq*wsq;
// if (val>0) {
if ((val>0)&&(val2>0)) {
return;
}
}
if (node->childs==NULL) {
for (int i=0; isize; ++i) {
int index = node->indices[i];
DistanceType dist = distance_(dataset_[index], vec, veclen_);
result.addPoint(dist, index);
}
}
else {
int* sort_indices = new int[branching_];
getCenterOrdering(node, vec, sort_indices);
for (int i=0; ichilds[sort_indices[i]],result,vec);
}
delete[] sort_indices;
}
}
/**
* Helper function.
*
* I computes the order in which to traverse the child nodes of a particular node.
*/
void getCenterOrdering(KMeansNodePtr node, const ElementType* q, int* sort_indices)
{
DistanceType* domain_distances = new DistanceType[branching_];
for (int i=0; ichilds[i]->pivot, veclen_);
int j=0;
while (domain_distances[j]j; --k) {
domain_distances[k] = domain_distances[k-1];
sort_indices[k] = sort_indices[k-1];
}
domain_distances[j] = dist;
sort_indices[j] = i;
}
delete[] domain_distances;
}
/**
* Method that computes the squared distance from the query point q
* from inside region with center c to the border between this
* region and the region with center p
*/
DistanceType getDistanceToBorder(DistanceType* p, DistanceType* c, DistanceType* q)
{
DistanceType sum = 0;
DistanceType sum2 = 0;
for (int i=0; ivariance*root->size;
while (clusterCount::max)();
int splitIndex = -1;
for (int i=0; ichilds != NULL) {
DistanceType variance = meanVariance - clusters[i]->variance*clusters[i]->size;
for (int j=0; jchilds[j]->variance*clusters[i]->childs[j]->size;
}
if (variance clusters_length) break;
meanVariance = minVariance;
// split node
KMeansNodePtr toSplit = clusters[splitIndex];
clusters[splitIndex] = toSplit->childs[0];
for (int i=1; ichilds[i];
}
}
varianceValue = meanVariance/root->size;
return clusterCount;
}
private:
/** The branching factor used in the hierarchical k-means clustering */
int branching_;
/** Maximum number of iterations to use when performing k-means clustering */
int iterations_;
/** Algorithm for choosing the cluster centers */
flann_centers_init_t centers_init_;
/**
* Cluster border index. This is used in the tree search phase when determining
* the closest cluster to explore next. A zero value takes into account only
* the cluster centres, a value greater then zero also take into account the size
* of the cluster.
*/
float cb_index_;
/**
* The dataset used by this index
*/
const Matrix dataset_;
/** Index parameters */
IndexParams index_params_;
/**
* Number of features in the dataset.
*/
size_t size_;
/**
* Length of each feature.
*/
size_t veclen_;
/**
* The root node in the tree.
*/
KMeansNodePtr root_;
/**
* Array of indices to vectors in the dataset.
*/
int* indices_;
/**
* The distance
*/
Distance distance_;
/**
* Pooled memory allocator.
*/
PooledAllocator pool_;
/**
* Memory occupied by the index.
*/
int memoryCounter_;
};
}
#endif //OPENCV_FLANN_KMEANS_INDEX_H_
