void Model::Train_em( const SampleSet& samples )
{
CvEM* model = (CvEM*)m_pModel;
CvEMParams* para = (CvEMParams*)m_trainPara;
para->nclusters = samples.Classes().size();
model->train(samples.Samples());
}
void Recoloring::TrainGMM(CvEM& source_model, Mat& source, Mat& source_mask) {
int src_samples_size = countNonZero(source_mask);
Mat source_samples(src_samples_size,3,CV_32FC1);
Mat source_32f;
//if(source.type() != CV_32F)
// source.convertTo(source_32f,CV_32F,1.0/255.0);
//else
source_32f = source;
int sample_count = 0;
for(int y=0;y<source.rows;y++) {
Vec3f* row = source_32f.ptr<Vec3f>(y);
uchar* mask_row = source_mask.ptr<uchar>(y);
for(int x=0;x<source.cols;x++) {
if(mask_row[x] > 0) {
source_samples.at<Vec3f>(sample_count++,0) = row[x];
}
}
}
source_model.clear();
CvEMParams ps(1/* = number of gaussians*/);
source_model.train(source_samples,Mat(),ps,NULL);
}
vector<int> Recoloring::MatchGaussians(CvEM& source_model, CvEM& target_model) {
int num_g = source_model.get_nclusters();
Mat sMu(source_model.get_means());
Mat tMu(target_model.get_means());
const CvMat** target_covs = target_model.get_covs();
const CvMat** source_covs = source_model.get_covs();
double best_dist = std::numeric_limits<double>::max();
vector<int> best_res(num_g);
vector<int> prmt(num_g);
for(int itr = 0; itr < 10; itr++) {
for(int i=0;i<num_g;i++) prmt[i] = i; //make a permutation
randShuffle(Mat(prmt));
//Greedy selection
vector<int> res(num_g);
vector<bool> taken(num_g);
for(int sg = 0; sg < num_g; sg++) {
double min_dist = std::numeric_limits<double>::max();
int minv = -1;
for(int tg = 0; tg < num_g; tg++) {
if(taken[tg]) continue;
//TODO: can save on re-calculation of pairs - calculate affinity matrix ahead
//double d = norm(sMu(Range(prmt[sg],prmt[sg]+1),Range(0,3)), tMu(Range(tg,tg+1),Range(0,3)));
//symmetric kullback-leibler
Mat diff = Mat(sMu(Range(prmt[sg],prmt[sg]+1),Range(0,3)) - tMu(Range(tg,tg+1),Range(0,3)));
Mat d = diff * Mat(Mat(source_covs[prmt[sg]]).inv() + Mat(target_covs[tg]).inv()) * diff.t();
Scalar tr = trace(Mat(
Mat(Mat(source_covs[prmt[sg]])*Mat(target_covs[tg])) +
Mat(Mat(target_covs[tg])*Mat(source_covs[prmt[sg]]).inv()) +
Mat(Mat::eye(3,3,CV_64FC1)*2)
));
double kl_dist = ((double*)d.data)[0] + tr[0];
if(kl_dist<min_dist) {
min_dist = kl_dist;
minv = tg;
}
}
res[prmt[sg]] = minv;
taken[minv] = true;
}
double dist = 0;
for(int i=0;i<num_g;i++) {
dist += norm(sMu(Range(prmt[i],prmt[i]+1),Range(0,3)),
tMu(Range(res[prmt[i]],res[prmt[i]]+1),Range(0,3)));
}
if(dist < best_dist) {
best_dist = dist;
best_res = res;
}
}
return best_res;
}
void Model::Predict_em( const SampleSet& samples, SampleSet& outError )
{
int true_resp = 0;
CvEM *model = (CvEM*)m_pModel;
for (int i = 0; i < samples.N(); i++)
{
float ret = model->predict(samples.GetSampleAt(i));
if (ret != samples.GetLabelAt(i))
{
outError.Add(samples.GetSampleAt(i), samples.GetLabelAt(i));
}
else
{
true_resp++;
}
}
printf("%d %d",samples.N(), true_resp);
}
float* pkmGaussianMixtureModel::getClusterMean(int clusterNum)
{
cout << "Variables: " << m_nVariables << endl;
float *returnedMeans = new float[m_nVariables];
if(bModeled)
{
CvEM *myModel = emModel + bestModel;
int numClusters = myModel->get_nclusters();
cout << "numClusters: " << numClusters << endl;
if (clusterNum < numClusters) {
const CvMat *modelMus = myModel->get_means();
if (modelMus != NULL) {
cout << "mus rows: " << modelMus->rows << " cols: " << modelMus->cols << endl;
for (int i = 0; i < m_nVariables; i++)
{
returnedMeans[i] = cvmGet(modelMus, clusterNum, i);
}
}
}
else
{
for (int i = 0; i < m_nVariables; i++)
{
returnedMeans[i] = 0;
}
}
}
else
{
for (int i = 0; i < m_nVariables; i++)
{
returnedMeans[i] = 0;
}
}
return returnedMeans;
}
int pkmGaussianMixtureModel::writeToFile(ofstream &fileStream, bool writeClusterNums, bool writeWeights, bool writeMeans, bool writeCovs, bool verbose)
{
if(!fileStream.is_open())
return -1;
// use the best-model
CvEM myModel = emModel[bestModel];
const CvMat **modelCovs = myModel.get_covs();
const CvMat *modelMus = myModel.get_means();
const CvMat *modelWeights = myModel.get_weights();
int numClusters = myModel.get_nclusters();
// output the total number of clusters
if(writeClusterNums)
{
if(verbose)
fileStream << "Number of Clusters: " << numClusters << "\n";
else
fileStream << numClusters << "\n";
}
if(writeWeights)
{
// output the weights of each cluster
if(verbose)
fileStream << "Weight of Clusters\n";
for (int k = 0; k < numClusters; k++)
{
const double *weightPtr = (const double *)(modelWeights->data.ptr + k*modelWeights->step);
double weight = weightPtr[0];
if(verbose)
fileStream << k << ": " << weight << "\n";
else
fileStream << weight << " ";
}
fileStream << "\n";
}
if(writeMeans)
{
// output the means of each cluster
if(verbose)
fileStream << "Means of Clusters\n";
for (int k = 0; k < numClusters; k++)
{
if(verbose)
fileStream << "Cluster " << k << ":\n";
for (int i = 0; i < m_nVariables; i++)
{
if(verbose)
fileStream << i << ": " << cvmGet(modelMus, k, i) << " ";
else
fileStream << cvmGet(modelMus, k, i) << " ";
}
fileStream << "\n";
}
}
if(writeCovs)
{
// output the covariances of each cluster
if(verbose)
fileStream << "Covariances of Clusters\n";
for (int k = 0; k < numClusters; k++)
{
const CvMat * covar = modelCovs[k];
if(verbose)
fileStream << "Cluster " << k << ":\n";
for (int i = 0; i < m_nVariables; i++)
{
for (int j = 0; j < m_nVariables; j++)
{
if(verbose)
fileStream << i << "," << j << ": " << cvmGet(covar, i, j) << " ";
else
fileStream << cvmGet(covar, i, j) << " ";
}
}
fileStream << "\n";
}
}
return 0;
}
/*void pkmGaussianMixtureModel::getLikelihood(int x, int y)
{
}
*/
void pkmGaussianMixtureModel::getLikelihoodMap(int rows, int cols, unsigned char *map, ofstream &filePtr, int widthstep)
{
if(widthstep == 0)
widthstep = cols;
if(!bModeled)
return;
CvEM myModel = emModel[bestModel];
const CvMat **modelCovs = myModel.get_covs();
const CvMat *modelMus = myModel.get_means();
const CvMat *modelWeights = myModel.get_weights();
int numClusters = myModel.get_nclusters();
CvMat *pts = cvCreateMat(m_nVariables, 1, CV_64FC1);
CvMat *mean = cvCreateMat(m_nVariables, 1, CV_64FC1);
double p = 0;
double weight;
double max = 0;
double prob;
filePtr << "clusters: " << numClusters << "\n";
filePtr << "likelihood: " << m_Likelihood << "\n";
filePtr << "BIC: " << m_BIC << "\n";
//printf("clusters: %d\n", numClusters);
float best_weight = 0;
bestCluster = 0;
for (int k = 0; k < numClusters; k++)
{
const CvMat * covar = modelCovs[k];
//printf("covar: (%f, %f, %f, %f)\n", covar->data.ptr[0], \
covar->data.ptr[1], \
covar->data.ptr[2], \
covar->data.ptr[3]);
//const double *meanVals = (const double *)(modelMus->data.ptr + k*modelWeights->step);
cvmSet(mean, 0, 0, cvmGet(modelMus, k, 0));
cvmSet(mean, 1, 0, cvmGet(modelMus, k, 1));
//const double *weightPtr = (const double *)(modelWeights->data.ptr + k*modelWeights->step);
//weight = weightPtr[k];
weight = cvmGet(modelWeights, 0, k);
if (best_weight < weight) {
best_weight = weight;
bestCluster = k;
}
filePtr << "mean: " << cvmGet(modelMus, k, 0)*(double)m_nScale << " " << cvmGet(modelMus, k, 1)*(double)m_nScale << "\n";
filePtr << "covar: " << cvmGet(covar, 0, 0) << "\n";
filePtr << "weight: " << weight << "\n";
for (int i = 0; i < rows; i++)
{
for (int j = 0; j < cols; j++)
{
cvmSet(pts, 0, 0, (double)j);
cvmSet(pts, 1, 0, (double)i);
prob = multinormalDistribution(pts, mean, covar);
map[j+i*widthstep] += (int)((weight * prob)*(double)(rows*cols));
}
}
}
cvReleaseMat(&mean);
cvReleaseMat(&pts);
}