static std::vector<double> computeKMeans(
KMdata & datapoints,
uint64_t const k,
uint64_t const runs,
bool const debug
)
{
KMterm const term(runs, 0, 0, 0, 0.10, 0.10, 3, 0.50, 10, 0.95);
datapoints.buildKcTree();
KMfilterCenters ctrs(k, datapoints);
KMlocalHybrid kmAlg(ctrs, term);
ctrs = kmAlg.execute();
std::vector<double> centrevector;
for ( uint64_t i = 0; i < k; ++i )
{
centrevector.push_back(ctrs[i][0]);
if ( debug )
std::cerr << "centre[" << i << "]=" << ctrs[i][0] << std::endl;
}
std::sort(centrevector.begin(),centrevector.end());
return centrevector;
}
void im_compute_bdd_bow(const IMbdd& bdd,
std::map <std::string, struct svm_problem>& peopleBOW){
std::string path2bdd(bdd.getFolder());
std::vector<std::string> activities = bdd.getActivities();
std::vector<std::string> people = bdd.getPeople();
int dim = bdd.getDim();
int maxPts = bdd.getMaxPts();
int k = bdd.getK();
for(std::vector<std::string>::iterator person = people.begin();
person != people.end();
++person){
int currentActivity = 1;
struct svm_problem svmPeopleBOW;
svmPeopleBOW.l = 0;
svmPeopleBOW.x = NULL;
svmPeopleBOW.y = NULL;
std::cout << "Computing the svmProblem of " << *person << std::endl;
for(std::vector<std::string>::iterator activity = activities.begin();
activity != activities.end();
++activity){
string rep(path2bdd + "/" + *person + "/" + *activity + "/fp");
DIR * repertoire = opendir(rep.c_str());
if (!repertoire){
std::cerr << "Impossible to open the feature points directory!" << std::endl;
exit(EXIT_FAILURE);
}
struct dirent * ent;
while ( (ent = readdir(repertoire)) != NULL){
std::string file = ent->d_name;
if(file.compare(".") != 0 && file.compare("..") != 0){
std::string path2FPs(rep + "/" + file);
KMdata dataPts(dim,maxPts);
int nPts = importSTIPs(path2FPs, dim, maxPts, &dataPts);
if(nPts != 0){
dataPts.setNPts(nPts);
dataPts.buildKcTree();
KMfilterCenters ctrs(k, dataPts);
importCenters(path2bdd + "/" + "training.means", dim, k, &ctrs);
// Only one BOW
struct svm_problem svmBow = computeBOW(currentActivity,
dataPts,
ctrs);
addBOW(svmBow.x[0], svmBow.y[0], svmPeopleBOW);
destroy_svm_problem(svmBow);
}
}
}
closedir(repertoire);
currentActivity++;
}
peopleBOW.insert(std::make_pair<std::string, struct svm_problem>((*person), svmPeopleBOW));
}
}
int
CVML_API
doKMeans(int dim, int k, int numofSamples, std::vector<std::vector<double>>& _samples,std::vector<std::vector<double>> & center)
{
typedef double T;
center.clear();
center.resize(k);
try{
// execution parameters (see KMterm.h and KMlocal.h)
KMterm term(100, 0, 0, 0, // run for 100 stages
0.10, 0.10, 3, // other typical parameter values
0.50, 10, 0.95);
//int dim = 1; // dimension
int nPts = numofSamples; // number of data points
KMdata dataPts(dim, nPts); // allocate data storage
//KMdata *dataPts = new KMdata(dim, nPts);
for ( unsigned int i = 0 ; i < numofSamples ; i++ ){
for(int j=0;i<dim;j++){
dataPts[i][j] = _samples[i][j];
}
}
dataPts.buildKcTree(); // build filtering structure
KMfilterCenters ctrs(k, dataPts); // allocate centers
// run the algorithm
//KMlocalLloyds kmAlg(ctrs, term); // repeated Lloyd's
// KMlocalSwap kmAlg(ctrs, term); // Swap heuristic
// KMlocalEZ_Hybrid kmAlg(ctrs, term); // EZ-Hybrid heuristic
KMlocalHybrid kmAlg(ctrs, term); // Hybrid heuristic
ctrs = kmAlg.execute(); // execute
for ( unsigned int i = 0 ; i < k ; i++ ){
for(int j=0;i<dim;j++){
T v = static_cast<T>(ctrs[i][j]);
center[i].push_back(v);
}
}
return 1;
}
catch(...)
{
//cvml::TRACK("doKMeans error \n");
return 0;
}
return 1;
}
Foam::pointField Foam::searchableBox::coordinates() const
{
pointField ctrs(6);
const pointField pts = treeBoundBox::points();
const faceList& fcs = treeBoundBox::faces;
forAll(fcs, i)
{
ctrs[i] = fcs[i].centre(pts);
}
Foam::pointField Foam::oldCyclicPolyPatch::calcFaceCentres
(
const UList<face>& faces,
const pointField& points
)
{
pointField ctrs(faces.size());
forAll(faces, facei)
{
ctrs[facei] = faces[facei].centre(points);
}
Foam::pointField Foam::perfectInterface::calcFaceCentres
(
const primitivePatch& pp
)
{
const pointField& points = pp.points();
pointField ctrs(pp.size());
forAll(ctrs, patchFaceI)
{
ctrs[patchFaceI] = pp[patchFaceI].centre(points);
}
/**
* \fn void createTrainingMeans(std::string stipFile, int dim, int maxPts, int k, std::string meansFile)
* \brief Import HOG and HOF from a file and compute KMeans algorithm to create
* the file training.means.
*
* \param[in] stipFile The file containing the STIPs.
* \param[in] dim Points and centers's dimension.
* \param[in] maxPts The maximum number of data we can compute
* \param[in] k The number of centers
* \param[out] meansFile The file in wich we will save the KMeans centers.
*/
void createTrainingMeans(std::string stipFile,
int dim,
int maxPts,
int k,
std::string meansFile
){
KMdata dataPts(dim,maxPts);
int nPts = importSTIPs(stipFile, dim, maxPts, &dataPts);
dataPts.setNPts(nPts);
dataPts.buildKcTree();
KMfilterCenters ctrs(k, dataPts);
int ic = 3;
kmIvanAlgorithm(ic, dim, dataPts, k, ctrs);
exportCenters(meansFile, dim, k, ctrs);
}
vector<DataPoint> SpeechKMeans::WeightedKMeans(vector<DataPoint> &points,
vector<double> &weights, int k) {
KMterm term(100, 0, 0, 0, // run for 100 stages
0.10, 0.10, 3, // other typical parameter values
0.50, 10, 0.95);
int dim = problems_.num_features(); // dimension
int nPts = points.size(); // number of data points
KMdata dataPts(dim, nPts); // allocate data storage
for (int p = 0; p < nPts; ++p) {
dataPts[p] = new double[dim];
for (int i = 0; i < dim; ++i) {
dataPts[p][i] = points[p][i];
}
}
//kmUniformPts(dataPts.getPts(), nPts, dim); // generate random points
dataPts.buildKcTree(); // build filtering structure
KMfilterCenters ctrs(k, dataPts); // allocate centers
// run the algorithm
//KMlocalLloyds kmAlg(ctrs, term); // repeated Lloyd's
// KMlocalSwap kmAlg(ctrs, term); // Swap heuristic
// KMlocalEZ_Hybrid kmAlg(ctrs, term); // EZ-Hybrid heuristic
KMlocalHybrid kmAlg(ctrs, term); // Hybrid heuristic
ctrs = kmAlg.execute(); // execute
// print number of stages
cout << "Number of stages: " << kmAlg.getTotalStages() << "\n";
// print average distortion
cout << "Average distortion: " << ctrs.getDist(false)/nPts << "\n";
ctrs.print(); // print final centers
cerr << "copying points" << endl;
vector<DataPoint> results(k);
for (int j = 0; j < k; ++j) {
results[j].resize(dim);
for (int i = 0; i < dim; ++i) {
results[j][i] = ctrs.getCtrPts()[j][i];
}
}
cerr << "done copying points" << endl;
return results;
}
int
CVML_API
doKMeans(int dim, int k, int numofSamples, const double * _samples, double*& center)
{
typedef double T;
if(center)
{
delete [] center;
center=0;
}
center = new T[k*dim];
memset( center,0,sizeof(T)*k*dim );
try{
// execution parameters (see KMterm.h and KMlocal.h)
KMterm term(100, 0, 0, 0, // run for 100 stages
0.10, 0.10, 3, // other typical parameter values
0.50, 10, 0.95);
//int dim = 1; // dimension
int nPts = numofSamples; // number of data points
KMdata dataPts(dim, nPts); // allocate data storage
//KMdata *dataPts = new KMdata(dim, nPts);
for ( unsigned int i = 0 ; i < numofSamples ; i++ ){
for(int j=0;i<dim;j++){
int idx = i * dim + j;
dataPts[i][j] = _samples[idx];
}
}
dataPts.buildKcTree(); // build filtering structure
KMfilterCenters ctrs(k, dataPts); // allocate centers
// run the algorithm
//KMlocalLloyds kmAlg(ctrs, term); // repeated Lloyd's
// KMlocalSwap kmAlg(ctrs, term); // Swap heuristic
// KMlocalEZ_Hybrid kmAlg(ctrs, term); // EZ-Hybrid heuristic
KMlocalHybrid kmAlg(ctrs, term); // Hybrid heuristic
ctrs = kmAlg.execute(); // execute
for ( unsigned int i = 0 ; i < k ; i++ ){
for(int j=0;i<dim;j++){
int idx = i * dim + j;
center[idx] = static_cast<T>(ctrs[i][j]);
}
}
return 1;
}
catch(...)
{
//cvml::TRACK("doKMeans error \n");
return 0;
}
return 1;
}
int im_create_specifics_training_means(IMbdd bdd,
const std::vector<std::string>& trainingPeople
//std::vector <std::string> rejects
){
std::string path2bdd(bdd.getFolder());
int dim = bdd.getDim();
int maxPts = bdd.getMaxPts();
std::vector <std::string> activities = bdd.getActivities();
int nr_class = activities.size();
int nr_people = trainingPeople.size();
// The total number of centers
int k = bdd.getK();
int subK = k/nr_class;
if(k%nr_class != 0){
std::cerr << "K is no divisible by the number of activities !!" << std::endl;
exit(EXIT_FAILURE);
}
double ***vDataPts = new double**[nr_class];
int nrFP[nr_class]; // number of feature points for each class
int currentActivity = 0;
// For each activity
for(std::vector<std::string>::iterator activity = activities.begin() ;
activity != activities.end() ;
++activity){
nrFP[currentActivity] = 0;
// We concatenate all the training people
int nrFPpP[nr_people]; // number of feature points per person
double*** activityDataPts = new double**[nr_people];
int currentPerson = 0;
for(std::vector<std::string>::const_iterator person = trainingPeople.begin() ;
person != trainingPeople.end() ;
++person){
nrFPpP[currentPerson] = 0;
std::string rep(path2bdd + "/" + *person + "/" + *activity);
DIR * repertoire = opendir(rep.c_str());
if (!repertoire){
std::cerr << "Impossible to open the feature points directory!" << std::endl;
exit(EXIT_FAILURE);
}
// Checking that the file concatenate.<activity>.fp exists
struct dirent * ent = readdir(repertoire);
std::string file(ent->d_name);
while (ent && (file.compare("concatenate." + *activity + ".fp")) != 0){
ent = readdir(repertoire);
file = ent->d_name;
}
if(!ent){
std::cerr << "No file concatenate.<activity>.fp" << std::endl;
exit(EXIT_FAILURE);
}
// Importing the feature points
std::string path2FP(rep + "/" + file);
KMdata kmData(dim,maxPts);
nrFPpP[currentPerson] = importSTIPs(path2FP, dim, maxPts, &kmData);
if(nrFPpP[currentPerson] != 0){
activityDataPts[currentPerson] = new double*[nrFPpP[currentPerson]];
for(int n = 0 ; n<nrFPpP[currentPerson] ; n++){
activityDataPts[currentPerson][n] = new double [dim];
for(int d = 0 ; d<dim ; d++)
activityDataPts[currentPerson][n][d] = kmData[n][d];
}
} // else the current person does not participate in this activity
nrFP[currentActivity] += nrFPpP[currentPerson];
currentPerson++;
} // ++person
// Saving people in vDataPts
vDataPts[currentActivity] = new double*[nrFP[currentActivity]];
int index=0;
for(int p=0 ; p<nr_people ; p++){
for(int fp=0 ; fp<nrFPpP[p] ; fp++){
vDataPts[currentActivity][index] = new double[dim];
for(int d=0 ; d<dim ; d++){
vDataPts[currentActivity][index][d] = activityDataPts[p][fp][d];
}
index++;
}
} // index must be equal to nrFP[currentActivity] - 1
// Deleting activityDataPts
for(int p=0 ; p<nr_people ; p++){
for(int fp=0 ; fp < nrFPpP[p] ; fp++)
delete[] activityDataPts[p][fp];
delete[] activityDataPts[p];
}
delete[] activityDataPts;
currentActivity++;
} // ++activity
// Total number of feature points
int ttFP = 0;
for(int a=0 ; a<nr_class ; a++){
ttFP += nrFP[a];
}
// Memory allocation of the centers
double** vCtrs = new double*[k];
for(int i=0 ; i<k ; i++){
vCtrs[i] = new double[dim];
}
// Doing the KMeans algorithm for each activities
int ic = 3; // the iteration coefficient (Ivan's algorithm)
int currCenter = 0;
for(int i=0 ; i<nr_class ; i++){
KMdata kmData(dim,nrFP[i]);
for(int n=0 ; n<nrFP[i]; n++)
for(int d=0 ; d<dim ; d++)
kmData[n][d] = vDataPts[i][n][d];
kmData.setNPts(nrFP[i]);
kmData.buildKcTree();
KMfilterCenters kmCtrs(subK,kmData);
kmIvanAlgorithm(ic, dim, kmData, subK, kmCtrs);
for(int n=0 ; n<subK ; n++){
for(int d=0 ; d<dim ; d++){
vCtrs[currCenter][d] = kmCtrs[n][d];
}
currCenter++;
}
}
std::cout << "Concatenate KMdata" << std::endl;
// Concatenate all the KMdata
/* In reality it is not necessary
but the objectif KMfilterCenters must be initialized with
the KMdata */
KMdata dataPts(dim,ttFP);
int nPts = 0;
for (int i=0 ; i<nr_class ; i++){
for(int n=0 ; n<nrFP[i]; n++){
for(int d=0 ; d<dim ; d++){
dataPts[nPts][d] = vDataPts[i][n][d];
}
nPts++;
}
}
// Releasing vDataPts
for(int i=0 ; i<nr_class ; i++){
for(int n=0 ; n<nrFP[i] ; n++)
delete [] vDataPts[i][n];
delete [] vDataPts[i];
}
delete[] vDataPts;
dataPts.buildKcTree();
// Returning the true centers
KMfilterCenters ctrs(k,dataPts);
for(int n=0 ; n<k ; n++){
for(int d=0 ; d<dim ; d++){
ctrs[n][d] = vCtrs[n][d];
}
}
// Releasing vCtrs
for(int i=0 ; i<k ; i++)
delete [] vCtrs[i];
delete[]vCtrs;
exportCenters(bdd.getFolder() + "/" + bdd.getKMeansFile(),
dim, k, ctrs);
return k;
}
/**
* \fn void predictActivity(std::string, std::string bddName, int maxPts)
* \brief Predict the activity done in a video with an existant trained BDD.
*
* \param[in] The path to the video to predict.
* \param[in] bddName The name of the BDD containing videos.
* \param[in] maxPts The maximum number of STIPs we can extract.
* \return The name of the activity.
*/
void predictActivity(std::string videoPath,
std::string bddName
){
std::string path2bdd("bdd/" + bddName);
// Loading parameters
IMbdd bdd(bddName,path2bdd);
bdd.load_bdd_configuration(path2bdd.c_str(),"imconfig.xml");
int scale_num = bdd.getScaleNum();
std::string descriptor = bdd.getDescriptor();
int dim = bdd.getDim();
int k = bdd.getK();
int maxPts = bdd.getMaxPts();
//double p = getTrainProbability(path2bdd);
// Computing feature points
KMdata dataPts(dim,maxPts);
int nPts = 0;
nPts = extract_feature_points(videoPath,
scale_num, descriptor, dim,
maxPts, dataPts);
if(nPts == 0){
std::cerr << "No activity detected !" << std::endl;
exit(EXIT_FAILURE);
}
std::cout << nPts << " vectors extracted..." << std::endl;
dataPts.setNPts(nPts);
dataPts.buildKcTree();
KMfilterCenters ctrs(k, dataPts);
importCenters(bdd.getFolder() + "/" + bdd.getKMeansFile(), dim, k, &ctrs);
std::cout << "KMeans centers imported..." << std::endl;
activitiesMap *am;
int nbActivities = mapActivities(path2bdd,&am);
struct svm_problem svmProblem = computeBOW(0,
dataPts,
ctrs);
double means[k], stand_devia[k];
load_gaussian_parameters(bdd, means, stand_devia);
// simple, gaussian, both, nothing
if(bdd.getNormalization().compare("") != 0)
bow_normalization(bdd,svmProblem);
std::cout << "Bag of words normalized..." << std::endl;
struct svm_model** pSVMModels = new svm_model*[nbActivities];
std::vector<std::string> modelFiles(bdd.getModelFiles());
int i=0;
for (std::vector<std::string>::iterator it = modelFiles.begin() ; it != modelFiles.end() ; ++it){
pSVMModels[i]= svm_load_model((*it).c_str());
i++;
}
std::cout << "SVM models imported..." << std::endl;
double probs[nbActivities];
double label = svm_predict_ovr_probs(pSVMModels,
svmProblem.x[0],
nbActivities,
probs,
2);
std::cerr<<"Probs: ";
for(int j=0 ; j<nbActivities ; j++){
std::cout << setw(2) << setiosflags(ios::fixed) << probs[j]*100<<" ";
}
destroy_svm_problem(svmProblem);
int index = searchMapIndex(label, am, nbActivities);
std::cout << "Activity predicted: ";
std::cout << am[index].activity << "(" << am[index].label << ")";
std::cout << std::endl;
for(int m=0 ; m<nbActivities ; m++){
svm_free_and_destroy_model(&pSVMModels[m]);
}
}
Foam::pointField Foam::searchableCylinder::coordinates() const
{
pointField ctrs(1, 0.5*(point1_ + point2_));
return ctrs;
}
void KMLProxy::run(Particles *initial_centers) {
IMP_INTERNAL_CHECK(is_init_,"The proxy was not initialized");
IMP_LOG(VERBOSE,"KMLProxy::run start \n");
//use the initial centers if provided
KMPointArray *kmc=nullptr;
if (initial_centers != nullptr) {
IMP_INTERNAL_CHECK(kcenters_ == initial_centers->size(),
"the number of initial points differs from the number of required"
<<" centers\n");
IMP_LOG(VERBOSE,"KMLProxy::run initial centers provided : \n");
kmc = allocate_points(kcenters_,atts_.size());
for (unsigned int i=0;i<kcenters_;i++){
Particle *cen=(*initial_centers)[i];
for(unsigned int j=0;j<atts_.size();j++) {
(*(*kmc)[i])[j]=cen->get_value(atts_[j]);
}
}
}
IMP_LOG(VERBOSE,"KMLProxy::run load initial guess \n");
//load the initail guess
KMFilterCenters ctrs(kcenters_, data_, kmc,damp_factor_);
//apply lloyd search
IMP_LOG(VERBOSE,"KMLProxy::run load lloyd \n");
lloyd_alg_ = new KMLocalSearchLloyd(&ctrs,&term_);
log_header();
IMP_CHECK_CODE(clock_t start = clock());
IMP_LOG(VERBOSE,"KMLProxy::run excute lloyd \n");
lloyd_alg_->execute();
IMP_LOG(VERBOSE,"KMLProxy::run analyse \n");
KMFilterCentersResults best_clusters = lloyd_alg_->get_best();
IMP_CHECK_CODE(Float exec_time = elapsed_time(start));
// print summary
IMP_LOG_WRITE(TERSE,log_summary(&best_clusters,exec_time));
IMP_LOG_WRITE(TERSE,best_clusters.show(IMP_STREAM));
IMP_INTERNAL_CHECK(kcenters_
== (unsigned int) best_clusters.get_number_of_centers(),
"The final number of centers does not match the requested one");
IMP_INTERNAL_CHECK (dim_ == (unsigned int) best_clusters.get_dim(),
"The dimension of the final clusters is wrong");
//TODO clear the centroids list
//set the centroids:
Particle *p;
IMP_LOG(VERBOSE,"KMLProxy::run load best results \n");
for (unsigned int ctr_ind = 0; ctr_ind < kcenters_; ctr_ind++) {
KMPoint *kmp = best_clusters[ctr_ind];
//create a new particle
p = new Particle(m_);
centroids_.push_back(p);
for (unsigned int att_ind = 0; att_ind < dim_; att_ind++) {
p->add_attribute(atts_[att_ind],(*kmp)[att_ind],false);
}
}
//set the assignment of particles to centers
//array of number of all points
//TODO - return this
IMP_LOG(VERBOSE,"KMLProxy::run get assignments \n");
const Ints *close_center = best_clusters.get_assignments();
IMP_LOG(VERBOSE,"KMLProxy::run get assignments 2\n");
for (int i=0;i<data_->get_number_of_points();i++) {
//std::cout<<"ps number i: " << i << " close center : "
//<< (*close_center)[i] << std::endl;
assignment_[ps_[i]]=(*close_center)[i];
}
}
//----------------------------------------------------------------------
// Main program
//----------------------------------------------------------------------
int nao_kmeans()
{
ifstream importMeans; // permettra d'utiliser les centres enregistrés dans l'étape de training
int nPts; // actual number of points
getArgs(); // read command-line arguments
term.setAbsMaxTotStage(stages); // set number of stages
// I- IMPORT STIPS
nPts = 0;
dim = STIPS_DIMENSION;
KMdata dataPts(dim, maxPts); // allocate data storage
if (dataIn != NULL){
laptevParser(*dataIn, dataPts, &nPts);
}
else{
perror("Pas de données à lire !!!");
return EXIT_FAILURE;
}
dataPts.setNPts(nPts); // set actual number of pts
dataPts.buildKcTree(); // build filtering structure
// II- IMPORT TRAINING CENTERS
KMfilterCenters ctrs(k, dataPts); // allocate centers
importMeans.open("/home/nao/data/activities_recognition/training.means", ios::in);
if(importMeans != NULL){
trainingMeansParser(importMeans,ctrs);
importMeans.close();
}
else{
cerr << "Error while importing Means" << endl;
return 2;
}
// III- GET ASSIGNMENTS
KMctrIdxArray closeCtr = new KMctrIdx[dataPts.getNPts()]; // dataPts = 1 label
double* sqDist = new double[dataPts.getNPts()];
ctrs.getAssignments(closeCtr, sqDist);
// IV- BAG OF WORDS STEP
// initialisation de l'histogramme
int* bowHistogram = NULL;
bowHistogram = new int[k];
for(int centre = 0; centre<k; centre++)
bowHistogram[centre]=0;
// remplissage de l'histogramme
for(int point = 0; point < nPts ; point++){
bowHistogram[closeCtr[point]]++;
}
delete closeCtr;
delete[] sqDist;
// exportation dans le fichier "testing.bow" sous un format pouvant être lu par libSVM
char* KMeansToBow = NULL;
KMeansToBow = new char[(256+1)];
sprintf(KMeansToBow,"%s","/home/nao/data/activities_recognition/stip"); //sprintf(KMeansToBow,"%s.bow",argv[1]);
ofstream testingBow(KMeansToBow, ios::out | ios::trunc); // ouverture en écriture avec effacement du fichier ouvert
if(testingBow){
testingBow << "0";
for(int centre = 0; centre<k ; centre++){
testingBow << " " << centre + 1 << ":" << bowHistogram[centre];
}
testingBow << endl;
testingBow.close();
}
else
cerr << "Impossible d'ouvrir le fichier erreur1" << endl;
// affichage
cout << "Bag Of Words histogram" << endl;
for(int centre = 0; centre<k; centre++){
cout << "Centre " << centre << ": ";
for (int i = 0; i<bowHistogram[centre]; i++)
cout << "*";
cout << endl;
}
delete[] bowHistogram;
//kmExit(0);
cout << "[kmeans.cpp]delete[]" << endl;
}
