Peter M. Groen / openbr

Commit 89c438fefc3a9174ecf07d7898f7ad1b583985b2

Authored by Josh Klontz
1 parent 85e03bed

progress on Bayesian quantization

Inline Side-by-side
Showing 2 changed files with 44 additions and 8 deletions
sdk/plugins/algorithms.cpp
  View file @89c438f
... ... @@ -42,7 +42,7 @@ class AlgorithmsInitializer : public Initializer
42 42 Globals->abbreviations.insert("OpenBR", "FaceRecognition");
43 43 Globals->abbreviations.insert("GenderEstimation", "GenderClassification");
44 44 Globals->abbreviations.insert("AgeEstimation", "AgeRegression");
45   - Globals->abbreviations.insert("FaceRecognitionHoG", "Open+Cvt(Gray)+Cascade(FrontalFace)+ASEFEyes+Affine(64,64,0.25,0.35)+Gradient+Bin(0,360,8,true)+Merge+Integral+IntegralSampler+RootNorm+ProductQuantization:ProductQuantization");
  45 + Globals->abbreviations.insert("FaceRecognitionHoG", "Open+Cvt(Gray)+Cascade(FrontalFace)+ASEFEyes+Affine(64,64,0.25,0.35)+Gradient+Bin(0,360,8,true)+Merge+Integral+IntegralSampler+RootNorm+ProductQuantization(2,true):ProductQuantization(true)");
46 46  
47 47 // Generic Image Processing
48 48 Globals->abbreviations.insert("SIFT", "Open+KeyPointDetector(SIFT)+KeyPointDescriptor(SIFT):KeyPointMatcher(BruteForce)");
... ...
sdk/plugins/quantize.cpp
  View file @89c438f
... ... @@ -14,6 +14,8 @@
14 14 * limitations under the License. *
15 15 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
16 16  
  17 +#include <QFutureSynchronizer>
  18 +#include <QtConcurrentRun>
17 19 #include <openbr_plugin.h>
18 20  
19 21 #include "core/opencvutils.h"
... ... @@ -131,7 +133,7 @@ class ProductQuantizationDistance : public Distance
131 133 const uchar *bData = b[i].data;
132 134 const float *lut = (const float*)ProductQuantizationLUTs[i].data;
133 135 for (int j=0; j<elements; j++)
134   - distance += lut[i*256*256 + aData[j]*256+bData[j]];
  136 + distance += lut[i*256*256 + aData[j]*256+bData[j]];
135 137 }
136 138 if (!bayesian) distance = -log(distance+1);
137 139 return distance;
... ... @@ -164,22 +166,56 @@ public:
164 166 }
165 167  
166 168 private:
  169 + static double likelihoodRatio(const QPair<int,int> &totals, const QList<int> &targets, const QList<int> &queries)
  170 + {
  171 + int positives = 0, negatives = 0;
  172 + foreach (int t, targets)
  173 + foreach (int q, queries)
  174 + if (t == q) positives++;
  175 + else negatives++;
  176 + return log(float(positives)/float(totals.first)) / (float(negatives)/float(totals.second));
  177 + }
  178 +
  179 + void _train(const Mat &data, const QPair<int,int> &totals, Mat &lut, int i, const QList<int> &templateLabels)
  180 + {
  181 + Mat labels, center;
  182 + kmeans(data.colRange(i*n,(i+1)*n), 256, labels, TermCriteria(TermCriteria::MAX_ITER, 10, 0), 3, KMEANS_PP_CENTERS, center);
  183 + QList<int> clusterLabels = OpenCVUtils::matrixToVector<int>(labels);
  184 + QHash< int, QList<int> > clusters; // QHash<clusterLabel, QList<templateLabel> >
  185 + for (int i=0; i<clusterLabels.size(); i++)
  186 + clusters[clusterLabels[i]].append(templateLabels[i]);
  187 +
  188 + for (int j=0; j<256; j++)
  189 + for (int k=0; k<256; k++)
  190 + lut.at<float>(i,j*256+k) = bayesian ? likelihoodRatio(totals, clusters[j], clusters[k]) :
  191 + norm(center.row(j), center.row(k), NORM_L2);
  192 + centers[i] = center;
  193 + }
  194 +
167 195 void train(const TemplateList &src)
168 196 {
169 197 Mat data = OpenCVUtils::toMat(src.data());
170 198 if (data.cols % n != 0) qFatal("Expected dimensionality to be divisible by n.");
  199 + const QList<int> templateLabels = src.labels<int>();
  200 + int totalPositives = 0, totalNegatives = 0;
  201 + for (int i=0; i<templateLabels.size(); i++)
  202 + for (int j=0; j<templateLabels.size(); j++)
  203 + if (templateLabels[i] == templateLabels[j]) totalPositives++;
  204 + else totalNegatives++;
  205 + QPair<int,int> totals(totalPositives, totalNegatives);
171 206  
172 207 Mat &lut = ProductQuantizationLUTs[index];
173 208 lut = Mat(data.cols/n, 256*256, CV_32FC1);
174 209  
  210 + for (int i=0; i<lut.rows; i++)
  211 + centers.append(Mat());
  212 +
  213 + QFutureSynchronizer<void> futures;
175 214 for (int i=0; i<lut.rows; i++) {
176   - Mat labels, center;
177   - kmeans(data.colRange(i*n,(i+1)*n), 256, labels, TermCriteria(TermCriteria::MAX_ITER, 10, 0), 3, KMEANS_PP_CENTERS, center);
178   - for (int j=0; j<256; j++)
179   - for (int k=0; k<256; k++)
180   - lut.at<float>(i,j*256+k) = norm(center.row(j), center.row(k), NORM_L2);
181   - centers.append(center);
  215 + if (Globals->parallelism) futures.addFuture(QtConcurrent::run(this, &ProductQuantizationTransform::_train, data, totals, lut, i, templateLabels));
  216 + else _train (data, totals, lut, i, templateLabels);
182 217 }
  218 + futures.waitForFinished();
183 219 }
184 220  
185 221 void project(const Template &src, Template &dst) const
... ...