Peter M. Groen / openbr

Commit 7a2db13ce795e7cca4c993c57d3ed3ca0b46c0df

Authored by Jordan Cheney
1 parent 98249fb4

No regression. Steps towards I/O.

Inline Side-by-side
Showing 7 changed files with 89 additions and 390 deletions
openbr/core/boost.cpp
  View file @7a2db13
... ... @@ -806,38 +806,6 @@ CvDTreeNode* CascadeBoostTree::predict( int sampleIdx ) const
806 806 return node;
807 807 }
808 808  
809   -<<<<<<< HEAD
810   -
811   -static void writeRecursive(FileStorage &fs, CvDTreeNode *node, int maxCatCount)
812   -{
813   - bool hasChildren = node->left ? true : false;
814   - fs << "hasChildren" << hasChildren;
815   -
816   - if (!hasChildren) // Write the leaf value
817   - fs << "value" << node->value; // value of the node. Only relevant for leaf nodes
818   - else { // Write the splitting information and then the children
819   - if (maxCatCount > 1) {
820   - fs << "subset" << "[:";
821   - for (int i = 0; i < ((maxCatCount + 31) / 32); i++)
822   - fs << node->split->subset[i]; // subset to split on (categorical features)
823   - fs << "]";
824   - } else {
825   - fs << "threshold" << node->split->ord.c; // threshold to split on (ordered features)
826   - }
827   -
828   - fs << "feature_idx" << node->split->var_idx; // feature idx of node
829   -
830   - fs << "left" << "{"; writeRecursive(fs, node->left, maxCatCount); fs << "}"; // write left child
831   - fs << "right" << "{"; writeRecursive(fs, node->right, maxCatCount); fs << "}"; // write right child
832   - }
833   -}
834   -
835   -void CascadeBoostTree::write(FileStorage &fs)
836   -{
837   - fs << "{";
838   - writeRecursive(fs, root, ((CascadeBoostTrainData*)data)->featureEvaluator->getMaxCatCount());
839   - fs << "}";
840   -}
841 809 /*
842 810 static void readRecursive(const FileNode &fn, CvDTreeNode *node, CvDTreeTrainData *data)
843 811 {
... ... @@ -881,8 +849,6 @@ void CascadeBoostTree::read(const FileNode &amp;fn, CvBoost* _ensemble, CvDTreeTrain
881 849 }
882 850 */
883 851  
884   -=======
885   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
886 852 void CascadeBoostTree::split_node_data( CvDTreeNode* node )
887 853 {
888 854 int n = node->sample_count, nl, nr, scount = data->sample_count;
... ... @@ -1138,11 +1104,7 @@ bool CascadeBoost::train( const FeatureEvaluator* _featureEvaluator,
1138 1104 break;
1139 1105 }
1140 1106  
1141   -<<<<<<< HEAD
1142 1107 classifiers.append(tree);
1143   -=======
1144   - trees.append(tree);
1145   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
1146 1108 cvSeqPush( weak, &tree );
1147 1109 update_weights( tree );
1148 1110 trim_weights();
... ... @@ -1470,7 +1432,3 @@ bool CascadeBoost::isErrDesired()
1470 1432  
1471 1433 return falseAlarm <= maxFalseAlarm;
1472 1434 }
1473   -<<<<<<< HEAD
1474   -
1475   -=======
1476   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
... ...
openbr/core/boost.h
  View file @7a2db13
... ... @@ -4,43 +4,6 @@
4 4 #include "ml.h"
5 5 #include <openbr/openbr_plugin.h>
6 6  
7   -#define CC_CASCADE_FILENAME "cascade.xml"
8   -#define CC_PARAMS_FILENAME "params.xml"
9   -
10   -#define CC_CASCADE_PARAMS "cascadeParams"
11   -#define CC_STAGE_TYPE "stageType"
12   -#define CC_FEATURE_TYPE "featureType"
13   -#define CC_HEIGHT "height"
14   -#define CC_WIDTH "width"
15   -
16   -#define CC_STAGE_NUM "stageNum"
17   -#define CC_STAGES "stages"
18   -#define CC_STAGE_PARAMS "stageParams"
19   -
20   -#define CC_BOOST "BOOST"
21   -#define CC_BOOST_TYPE "boostType"
22   -#define CC_DISCRETE_BOOST "DAB"
23   -#define CC_REAL_BOOST "RAB"
24   -#define CC_LOGIT_BOOST "LB"
25   -#define CC_GENTLE_BOOST "GAB"
26   -#define CC_MINHITRATE "minHitRate"
27   -#define CC_MAXFALSEALARM "maxFalseAlarm"
28   -#define CC_TRIM_RATE "weightTrimRate"
29   -#define CC_MAX_DEPTH "maxDeptrh"
30   -#define CC_WEAK_COUNT "maxWeakCount"
31   -#define CC_STAGE_THRESHOLD "stageThreshold"
32   -#define CC_WEAK_CLASSIFIERS "weakClassifiers"
33   -#define CC_INTERNAL_NODES "internalNodes"
34   -#define CC_LEAF_VALUES "leafValues"
35   -
36   -#define CC_FEATURES "features"
37   -#define CC_FEATURE_PARAMS "featureParams"
38   -#define CC_MAX_CAT_COUNT "maxCatCount"
39   -#define CC_FEATURE_SIZE "featSize"
40   -
41   -#define CC_LBP "LBP"
42   -#define CC_RECT "rect"
43   -
44 7 #ifdef _WIN32
45 8 #define TIME( arg ) (((double) clock()) / CLOCKS_PER_SEC)
46 9 #else
... ... @@ -125,22 +88,14 @@ public:
125 88 virtual float predict( int sampleIdx, bool returnSum = false ) const;
126 89  
127 90 float getThreshold() const { return threshold; }
128   -<<<<<<< HEAD
129   - QList<CvBoostTree*> getClassifiers() const { return classifiers; }
130   -=======
131   - const QList<CvBoostTree*> getTrees() const { return trees; }
132   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
  91 + QList<CvBoostTree *> getClassifers() const { return classifiers; }
133 92  
134 93 protected:
135 94 virtual bool set_params(const CvBoostParams& _params);
136 95 virtual void update_weights(CvBoostTree* tree);
137 96 virtual bool isErrDesired();
138 97  
139   -<<<<<<< HEAD
140   - QList<CvBoostTree*> classifiers;
141   -=======
142   - QList<CvBoostTree*> trees;
143   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
  98 + QList<CvBoostTree *> classifiers;
144 99  
145 100 float threshold;
146 101 float minHitRate, maxFalseAlarm;
... ...
openbr/core/cascade.cpp
  View file @7a2db13
... ... @@ -124,9 +124,10 @@ void br::groupRectangles(vector&lt;Rect&gt;&amp; rectList, vector&lt;int&gt;&amp; rejectLevels, vect
124 124 static void loadRecursive(const FileNode &fn, _CascadeClassifier::Node *node, int maxCatCount)
125 125 {
126 126 bool hasChildren = (int)fn["hasChildren"];
127   -
128   - if (hasChildren) {
129   - if (maxCatCount > 1) {
  127 + if (!hasChildren)
  128 + node->value = (float)fn["value"];
  129 + else {
  130 + if (maxCatCount > 0) {
130 131 FileNode subset_fn = fn["subset"];
131 132 for (FileNodeIterator subset_it = subset_fn.begin(); subset_it != subset_fn.end(); ++subset_it)
132 133 node->subset.append((int)*subset_it);
... ... @@ -134,14 +135,11 @@ static void loadRecursive(const FileNode &amp;fn, _CascadeClassifier::Node *node, in
134 135 node->threshold = (float)fn["threshold"];
135 136 }
136 137  
137   - node->featureIdx = (int)fn["feature_idx"];
  138 + node->featureIdx = (int)fn["featureIdx"];
138 139  
139   - node->left = new _CascadeClassifier::Node;
  140 + node->left = new _CascadeClassifier::Node; node->right = new _CascadeClassifier::Node;
140 141 loadRecursive(fn["left"], node->left, maxCatCount);
141   - node->right = new _CascadeClassifier::Node;
142 142 loadRecursive(fn["right"], node->right, maxCatCount);
143   - } else {
144   - node->value = (float)fn["value"];
145 143 }
146 144 }
147 145  
... ... @@ -159,7 +157,6 @@ bool _CascadeClassifier::load(const string&amp; filename)
159 157  
160 158 // load stages
161 159 FileNode stages_fn = root["stages"];
162   -
163 160 if( stages_fn.empty() )
164 161 return false;
165 162  
... ... @@ -170,7 +167,6 @@ bool _CascadeClassifier::load(const string&amp; filename)
170 167 stage.threshold = (float)stage_fn["stageThreshold"] - THRESHOLD_EPS;
171 168  
172 169 FileNode nodes_fn = stage_fn["weakClassifiers"];
173   -
174 170 if(nodes_fn.empty())
175 171 return false;
176 172  
... ...
openbr/plugins/classification/boostedforest.cpp
  View file @7a2db13
1 1 #include <openbr/plugins/openbr_internal.h>
2 2 #include <openbr/core/boost.h>
3 3  
  4 +#define THRESHOLD_EPS 1e-5
  5 +
4 6 using namespace cv;
5 7  
6 8 namespace br
... ... @@ -8,63 +10,33 @@ namespace br
8 10  
9 11 struct Node
10 12 {
11   - Node() : left(NULL), right(NULL) {}
12   -
13   -<<<<<<< HEAD
14   - float value;
  13 + float value; // for leaf nodes
15 14  
16   - float threshold; // For ordered features
17   - QList<int> subset; // For categorical features
  15 + float threshold; // for ordered features
  16 + QList<int> subset; // for categorical features
18 17 int featureIdx;
19 18  
20   -=======
21   - int featureIdx;
22   - float threshold; // for ordered features only
23   - QList<int> subset; // for categorical features only
24   - float value; // for leaf nodes only
25   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
26   - Node *left;
27   - Node *right;
  19 + Node *left, *right;
28 20 };
29 21  
30   -<<<<<<< HEAD
31 22 static void buildTreeRecursive(Node *node, const CvDTreeNode *cv_node, int maxCatCount)
32 23 {
33   - if (!cv_node->left) // Write the leaf value
34   - node->value = cv_node->value; // value of the node. Only relevant for leaf nodes
35   - else { // Write the splitting information and then the children
36   - if (maxCatCount > 1)
37   - for (int i = 0; i < ((maxCatCount + 31) / 32); i++)
38   - node->subset.append(cv_node->split->subset[i]); // subset to split on (categorical features)
  24 + if (!cv_node->left) {
  25 + node->value = cv_node->value;
  26 +
  27 + node->left = node->right = NULL;
  28 + } else {
  29 + if (maxCatCount > 0)
  30 + for (int i = 0; i < (maxCatCount + 31)/32; i++)
  31 + node->subset.append(cv_node->split->subset[i]);
39 32 else
40   - node->threshold = cv_node->split->ord.c; // threshold to split on (ordered features)
  33 + node->threshold = cv_node->split->ord.c;
41 34  
42   - node->featureIdx = cv_node->split->var_idx; // feature idx of node
  35 + node->featureIdx = cv_node->split->var_idx;
43 36  
44   - node->left = new Node;
  37 + node->left = new Node; node->right = new Node;
45 38 buildTreeRecursive(node->left, cv_node->left, maxCatCount);
46   - node->right = new Node;
47 39 buildTreeRecursive(node->right, cv_node->right, maxCatCount);
48   -=======
49   -static void buildTreeRecursive(Node *node, const CvDTreeNode *tree_node, int maxCatCount)
50   -{
51   - if (tree_node->left) {
52   - if (maxCatCount > 1) {
53   - for (int i = 0; i < (maxCatCount + 31)/32; i++)
54   - node->subset.append(tree_node->split->subset[i]);
55   - } else {
56   - node->threshold = tree_node->split->ord.c;
57   - }
58   -
59   - node->featureIdx = tree_node->split->var_idx;
60   -
61   - node->left = new Node;
62   - buildTreeRecursive(node->left, tree_node->left, maxCatCount);
63   - node->right = new Node;
64   - buildTreeRecursive(node->right, tree_node->right, maxCatCount);
65   - } else {
66   - node->value = tree_node->value;
67   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
68 40 }
69 41 }
70 42  
... ... @@ -73,58 +45,25 @@ static void writeRecursive(FileStorage &amp;fs, const Node *node, int maxCatCount)
73 45 bool hasChildren = node->left ? true : false;
74 46 fs << "hasChildren" << hasChildren;
75 47  
76   - if (!hasChildren) // Write the leaf value
77   -<<<<<<< HEAD
78   - fs << "value" << node->value; // value of the node. Only relevant for leaf nodes
79   - else { // Write the splitting information and then the children
80   - if (maxCatCount > 1) {
81   - fs << "subset" << "[:";
82   -=======
83   - fs << "value" << node->value; // value of the node.
84   - else { // Write the splitting information and then the children
85   - if (maxCatCount > 1) {
  48 + if (!hasChildren)
  49 + fs << "value" << node->value;
  50 + else {
  51 + if (maxCatCount > 0) {
86 52 fs << "subset" << "[";
87   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
88   - for (int i = 0; i < ((maxCatCount + 31) / 32); i++)
89   - fs << node->subset[i]; // subset to split on (categorical features)
  53 + for (int i = 0; i < (maxCatCount + 31)/32; i++)
  54 + fs << node->subset[i];
90 55 fs << "]";
91 56 } else {
92   - fs << "threshold" << node->threshold; // threshold to split on (ordered features)
93   - }
94   -
95   - fs << "feature_idx" << node->featureIdx; // feature idx of node
96   -
97   - fs << "left" << "{"; writeRecursive(fs, node->left, maxCatCount); fs << "}"; // write left child
98   - fs << "right" << "{"; writeRecursive(fs, node->right, maxCatCount); fs << "}"; // write right child
99   - }
100   -}
101   -
102   -<<<<<<< HEAD
103   -=======
104   -static void readRecursive(const FileNode &fn, Node *node, int maxCatCount)
105   -{
106   - bool hasChildren = (int)fn["hasChildren"];
107   - if (!hasChildren) {
108   - node->value = (float)fn["value"];
109   - } else {
110   - if (maxCatCount > 1) {
111   - FileNode subset_fn = fn["subset"];
112   - for (FileNodeIterator subset_it = subset_fn.begin(); subset_it != subset_fn.end(); ++subset_it)
113   - node->subset.append((int)*subset_it);
114   - } else {
115   - node->threshold = (float)fn["threshold"];
  57 + fs << "threshold" << node->threshold;
116 58 }
117 59  
118   - node->featureIdx = (int)fn["feature_idx"];
  60 + fs << "featureIdx" << node->featureIdx;
119 61  
120   - node->left = new Node;
121   - readRecursive(fn["left"], node->left, maxCatCount);
122   - node->right = new Node;
123   - readRecursive(fn["right"], node->right, maxCatCount);
  62 + fs << "left" << "{"; writeRecursive(fs, node->left, maxCatCount); fs << "}";
  63 + fs << "right" << "{"; writeRecursive(fs, node->right, maxCatCount); fs << "}";
124 64 }
125 65 }
126 66  
127   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
128 67 class BoostedForestClassifier : public Classifier
129 68 {
130 69 Q_OBJECT
... ... @@ -143,7 +82,7 @@ class BoostedForestClassifier : public Classifier
143 82 BR_PROPERTY(int, maxDepth, 1)
144 83 BR_PROPERTY(int, maxWeakCount, 100)
145 84  
146   - QList<Node*> weakClassifiers;
  85 + QList<Node*> classifiers;
147 86 float threshold;
148 87  
149 88 void train(const QList<Mat> &images, const QList<float> &labels)
... ... @@ -157,65 +96,39 @@ class BoostedForestClassifier : public Classifier
157 96 featureEvaluator.setImage(images[i], labels[i], i);
158 97  
159 98 CascadeBoost boost;
160   - boost.train(&featureEvaluator, images.size(), 2048, 2048, params);
161   -<<<<<<< HEAD
162   -
163   - threshold = boost.getThreshold();
164   -
165   - foreach (const CvBoostTree *tree, boost.getClassifiers()) {
166   -=======
  99 + boost.train(&featureEvaluator, images.size(), 1024, 1024, params);
167 100  
168   - // Convert into simpler, cleaner cascade after training
169 101 threshold = boost.getThreshold();
170 102  
171   - foreach (const CvBoostTree *tree, boost.getTrees()) {
172   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
  103 + foreach (const CvBoostTree *classifier, boost.getClassifers()) {
173 104 Node *root = new Node;
174   - buildTreeRecursive(root, tree->get_root(), representation->maxCatCount());
175   - weakClassifiers.append(root);
  105 + buildTreeRecursive(root, classifier->get_root(), representation->maxCatCount());
  106 + classifiers.append(root);
176 107 }
177 108 }
178 109  
179   - float classify(const Mat &image) const
  110 + float classify(const Mat &_image) const
180 111 {
181   -<<<<<<< HEAD
182   - float sum = 0;
183   - foreach (const Node *root, weakClassifiers) {
184   - const Node *node = root;
185   -
186   - while (node->left) {
187   - if (representation->maxCatCount() > 1) {
188   - int c = (int)representation->evaluate(image, node->featureIdx);
189   - node = (node->subset[c >> 5] & (1 << (c & 31))) ? node->left : node->right;
190   - } else {
191   - float val = representation->evaluate(image, node->featureIdx);
192   -=======
193   - Mat pp;
194   - representation->preprocess(image, pp);
  112 + Mat image;
  113 + representation->preprocess(_image, image);
195 114  
196 115 float sum = 0;
  116 + for (int i = 0; i < classifiers.size(); i++) {
  117 + Node *node = classifiers[i];
197 118  
198   - foreach (const Node *node, weakClassifiers) {
199 119 while (node->left) {
200 120 if (representation->maxCatCount() > 1) {
201   - int c = (int)representation->evaluate(pp, node->featureIdx);
202   - node = (node->subset[c >> 5] & (1 << (c & 31))) ? node->left : node->right;
  121 + int c = (int)representation->evaluate(image, node->featureIdx);
  122 + node = (2*((node->subset[c >> 5] & (1 << (c & 31))) == 0) - 1) < 0 ? node->left : node->right;
203 123 } else {
204   - double val = representation->evaluate(pp, node->featureIdx);
205   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
206   - node = val < node->threshold ? node->left : node->right;
  124 + double val = representation->evaluate(image, node->featureIdx);
  125 + node = val <= node->threshold ? node->left : node->right;
207 126 }
208 127 }
209 128 sum += node->value;
210 129 }
211   -<<<<<<< HEAD
212   - return sum < threshold - FLT_EPSILON ? 0.0 : 1.0;
213   -=======
214   -
215   - if (sum < threshold)
216   - return 0.0f; //-std::abs(sum);
217   - return 1.0f; //std::abs(sum);
218   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
  130 +
  131 + return sum < threshold - THRESHOLD_EPS ? 0.0f : 1.0f;
219 132 }
220 133  
221 134 int numFeatures() const
... ... @@ -235,38 +148,15 @@ class BoostedForestClassifier : public Classifier
235 148  
236 149 void write(FileStorage &fs) const
237 150 {
238   -<<<<<<< HEAD
239   - fs << "weakCount" << weakClassifiers.size();
240   -=======
241   - fs << "numWeak" << weakClassifiers.size();
242   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
243 151 fs << "stageThreshold" << threshold;
  152 + fs << "weakSize" << classifiers.size();
244 153 fs << "weakClassifiers" << "[";
245   - foreach (const Node *root, weakClassifiers) {
  154 + foreach (const Node *root, classifiers) {
246 155 fs << "{";
247 156 writeRecursive(fs, root, representation->maxCatCount());
248 157 fs << "}";
249 158 }
250 159 fs << "]";
251   -<<<<<<< HEAD
252   -=======
253   - }
254   -
255   - void read(const FileNode &node)
256   - {
257   - weakClassifiers.reserve((int)node["numWeak"]);
258   - threshold = (float)node["stageThreshold"];
259   -
260   - FileNode weaks_fn = node["weakClassifiers"];
261   - for (FileNodeIterator weaks_it = weaks_fn.begin(); weaks_it != weaks_fn.end(); ++weaks_it) {
262   - FileNode weak_fn = *weaks_it;
263   -
264   - Node *root = new Node;
265   - readRecursive(weak_fn, root, representation->maxCatCount());
266   -
267   - weakClassifiers.append(root);
268   - }
269   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
270 160 }
271 161 };
272 162  
... ...
openbr/plugins/classification/cascade.cpp
  View file @7a2db13
1 1 #include <opencv2/imgproc/imgproc.hpp>
2 2  
3 3 #include <openbr/plugins/openbr_internal.h>
4   -#include <openbr/core/boost.h>
5 4  
6 5 using namespace cv;
7 6  
... ... @@ -112,14 +111,12 @@ class CascadeClassifier : public Classifier
112 111 Q_PROPERTY(int numPos READ get_numPos WRITE set_numPos RESET reset_numPos STORED false)
113 112 Q_PROPERTY(int numNegs READ get_numNegs WRITE set_numNegs RESET reset_numNegs STORED false)
114 113 Q_PROPERTY(float maxFAR READ get_maxFAR WRITE set_maxFAR RESET reset_maxFAR STORED false)
115   - Q_PROPERTY(bool ROCMode READ get_ROCMode WRITE set_ROCMode RESET reset_ROCMode STORED false)
116 114  
117 115 BR_PROPERTY(QString, stageDescription, "")
118 116 BR_PROPERTY(int, numStages, 20)
119 117 BR_PROPERTY(int, numPos, 1000)
120 118 BR_PROPERTY(int, numNegs, 1000)
121 119 BR_PROPERTY(float, maxFAR, pow(0.5, numStages))
122   - BR_PROPERTY(bool, ROCMode, false)
123 120  
124 121 QList<Classifier *> stages;
125 122  
... ... @@ -158,21 +155,9 @@ class CascadeClassifier : public Classifier
158 155 float classify(const Mat &image) const
159 156 {
160 157 foreach (const Classifier *stage, stages)
161   - if (stage->classify(image) == 0)
  158 + if (stage->classify(image) == 0.0f)
162 159 return 0.0f;
163 160 return 1.0f;
164   -
165   - /*if (stages.size() == 0) // special case for empty cascade
166   - return 1.0f;
167   -
168   - float result = 0.0f;
169   - for (int stageIdx = 0; stageIdx < stages.size(); stageIdx++) {
170   - result = stages[stageIdx]->classify(image);
171   -
172   - if (result < 0)
173   - return stageIdx > (stages.size() - 4) ? stageIdx * result : 0.0f;
174   - }
175   - return std::abs(stages.size() * result);*/
176 161 }
177 162  
178 163 int numFeatures() const
... ... @@ -192,13 +177,7 @@ class CascadeClassifier : public Classifier
192 177  
193 178 void write(FileStorage &fs) const
194 179 {
195   -<<<<<<< HEAD
196   - fs << "stageCount" << stages.size();
197   -
198 180 fs << "stages" << "[";
199   -=======
200   - fs << CC_STAGES << "[";
201   ->>>>>>> 4fab7f69ddc82d6ba40a73fc6233e3cc9871473e
202 181 foreach (const Classifier *stage, stages) {
203 182 fs << "{";
204 183 stage->write(fs);
... ... @@ -217,7 +196,7 @@ private:
217 196 if (!imgHandler.getPos(pos))
218 197 qFatal("Cannot get another positive sample!");
219 198  
220   - if (classify(pos) > 0.0f) {
  199 + if (classify(pos) == 1.0f) {
221 200 printf("POS current samples: %d\r", images.size());
222 201 images.append(pos);
223 202 labels.append(1.0f);
... ... @@ -233,7 +212,7 @@ private:
233 212 if (!imgHandler.getNeg(neg))
234 213 qFatal("Cannot get another negative sample!");
235 214  
236   - if (classify(neg) > 0.0f) {
  215 + if (classify(neg) == 1.0f) {
237 216 printf("NEG current samples: %d\r", images.size() - posCount);
238 217 images.append(neg);
239 218 labels.append(0.0f);
... ...
openbr/plugins/imgproc/slidingwindow.cpp
  View file @7a2db13
... ... @@ -21,8 +21,6 @@
21 21 #include <openbr/core/qtutils.h>
22 22  
23 23 #include <opencv2/highgui/highgui.hpp>
24   -#include <opencv2/imgproc/imgproc.hpp>
25   -#include <opencv2/objdetect/objdetect.hpp>
26 24  
27 25 using namespace cv;
28 26  
... ... @@ -41,23 +39,11 @@ class SlidingWindowTransform : public Transform
41 39 Q_OBJECT
42 40  
43 41 Q_PROPERTY(br::Classifier *classifier READ get_classifier WRITE set_classifier RESET reset_classifier STORED false)
44   - Q_PROPERTY(int minSize READ get_minSize WRITE set_minSize RESET reset_minSize STORED false)
45   - Q_PROPERTY(int maxSize READ get_maxSize WRITE set_maxSize RESET reset_maxSize STORED false)
46   - Q_PROPERTY(float scaleFactor READ get_scaleFactor WRITE set_scaleFactor RESET reset_scaleFactor STORED false)
47   - Q_PROPERTY(int minNeighbors READ get_minNeighbors WRITE set_minNeighbors RESET reset_minNeighbors STORED false)
48   - Q_PROPERTY(float eps READ get_eps WRITE set_eps RESET reset_eps STORED false)
49   -
50 42 Q_PROPERTY(QString cascadeDir READ get_cascadeDir WRITE set_cascadeDir RESET reset_cascadeDir STORED false)
51 43 Q_PROPERTY(QString vecFile READ get_vecFile WRITE set_vecFile RESET reset_vecFile STORED false)
52 44 Q_PROPERTY(QString negFile READ get_negFile WRITE set_negFile RESET reset_negFile STORED false)
53 45  
54 46 BR_PROPERTY(br::Classifier *, classifier, NULL)
55   - BR_PROPERTY(int, minSize, 24)
56   - BR_PROPERTY(int, maxSize, -1)
57   - BR_PROPERTY(float, scaleFactor, 1.2)
58   - BR_PROPERTY(int, minNeighbors, 5)
59   - BR_PROPERTY(float, eps, 0.2)
60   -
61 47 BR_PROPERTY(QString, cascadeDir, "")
62 48 BR_PROPERTY(QString, vecFile, "vec.vec")
63 49 BR_PROPERTY(QString, negFile, "neg.txt")
... ... @@ -135,11 +121,6 @@ class SlidingWindowTransform : public Transform
135 121 return negs;
136 122 }
137 123  
138   - void init()
139   - {
140   - cascadeDir = Globals->sdkPath + "/share/openbr/models/openbrcascades/" + cascadeDir;
141   - }
142   -
143 124 void train(const TemplateList &_data)
144 125 {
145 126 (void)_data;
... ... @@ -159,109 +140,34 @@ class SlidingWindowTransform : public Transform
159 140 }
160 141  
161 142 classifier->train(images, labels);
  143 + }
  144 +
  145 + void project(const Template &src, Template &dst) const
  146 + {
  147 + (void)src; (void)dst;
  148 + }
  149 +
  150 + void load(QDataStream &stream)
  151 + {
  152 + (void) stream;
  153 + return;
  154 + }
162 155  
163   - // save the cascade
164   - std::string filename = cascadeDir.toStdString() + "/cascade.xml";
165   - FileStorage fs(filename, FileStorage::WRITE );
  156 + void store(QDataStream &stream) const
  157 + {
  158 + (void) stream;
166 159  
  160 + QString filename = Globals->sdkPath + "/share/openbr/models/openbrcascades/" + cascadeDir + "/cascade.xml";
  161 + FileStorage fs(filename.toStdString(), FileStorage::WRITE);
167 162 if ( !fs.isOpened() )
168 163 return;
169 164  
170   - fs << FileStorage::getDefaultObjectName(filename) << "{";
  165 + fs << FileStorage::getDefaultObjectName(filename.toStdString()) << "{";
171 166  
172 167 classifier->write(fs);
173 168  
174 169 fs << "}";
175 170 }
176   -
177   - void project(const Template &src, Template &dst) const
178   - {
179   - TemplateList temp;
180   - project(TemplateList() << src, temp);
181   - if (!temp.isEmpty()) dst = temp.first();
182   - }
183   -
184   - void project(const TemplateList &src, TemplateList &dst) const
185   - {
186   - foreach (const Template &t, src) {
187   - const bool enrollAll = t.file.getBool("enrollAll");
188   -
189   - // Mirror the behavior of ExpandTransform in the special case
190   - // of an empty template.
191   - if (t.empty() && !enrollAll) {
192   - dst.append(t);
193   - continue;
194   - }
195   -
196   - for (int i = 0; i < t.size(); i++) {
197   - Mat image;
198   - OpenCVUtils::cvtUChar(t[i], image);
199   -
200   - std::vector<Rect> rects;
201   - std::vector<int> rejectLevels;
202   - std::vector<double> levelWeights;
203   -
204   - Size minObjectSize(minSize, minSize);
205   - Size maxObjectSize(maxSize, maxSize);
206   - if (maxObjectSize.height == 0 || maxObjectSize.width == 0)
207   - maxObjectSize = image.size();
208   -
209   - Mat imageBuffer(image.rows + 1, image.cols + 1, CV_8U);
210   -
211   - for (double factor = 1; ; factor *= scaleFactor) {
212   - Size originalWindowSize = classifier->windowSize();
213   -
214   - Size windowSize(cvRound(originalWindowSize.width*factor), cvRound(originalWindowSize.height*factor) );
215   - Size scaledImageSize(cvRound(image.cols/factor ), cvRound(image.rows/factor));
216   - Size processingRectSize(scaledImageSize.width - originalWindowSize.width, scaledImageSize.height - originalWindowSize.height);
217   -
218   - if (processingRectSize.width <= 0 || processingRectSize.height <= 0)
219   - break;
220   - if (windowSize.width > maxObjectSize.width || windowSize.height > maxObjectSize.height)
221   - break;
222   - if (windowSize.width < minObjectSize.width || windowSize.height < minObjectSize.height)
223   - continue;
224   -
225   - Mat scaledImage(scaledImageSize, CV_8U, imageBuffer.data);
226   - resize(image, scaledImage, scaledImageSize, 0, 0, CV_INTER_LINEAR);
227   -
228   - int yStep = factor > 2. ? 1 : 2;
229   - for (int y = 0; y < processingRectSize.height; y += yStep) {
230   - for (int x = 0; x < processingRectSize.width; x += yStep) {
231   - Mat window = scaledImage(Rect(Point(x, y), classifier->windowSize())).clone();
232   -
233   - float result = classifier->classify(window);
234   -
235   - if (result > 0) {
236   - rects.push_back(Rect(cvRound(x*factor), cvRound(y*factor), windowSize.width, windowSize.height));
237   - rejectLevels.push_back(1);
238   - levelWeights.push_back(result);
239   - }
240   - if (result == 0)
241   - x += yStep;
242   - }
243   - }
244   - }
245   -
246   - groupRectangles(rects, rejectLevels, levelWeights, minNeighbors, eps);
247   -
248   - if (!enrollAll && rects.empty())
249   - rects.push_back(Rect(0, 0, image.cols, image.rows));
250   -
251   - for (size_t j = 0; j < rects.size(); j++) {
252   - Template u(t.file, image);
253   - if (rejectLevels.size() > j)
254   - u.file.set("Confidence", rejectLevels[j]*levelWeights[j]);
255   - else
256   - u.file.set("Confidence", 1);
257   - const QRectF rect = OpenCVUtils::fromRect(rects[j]);
258   - u.file.appendRect(rect);
259   - u.file.set("Face", rect);
260   - dst.append(u);
261   - }
262   - }
263   - }
264   - }
265 171 };
266 172  
267 173 BR_REGISTER(Transform, SlidingWindowTransform)
... ...
openbr/plugins/representation/mblbp.cpp
  View file @7a2db13
... ... @@ -58,6 +58,7 @@ class MBLBPRepresentation : public Representation
58 58 return result;
59 59 }
60 60  
  61 + void write(FileStorage &fs, const Mat &featureMap);
61 62 int numFeatures() const { return features.size(); }
62 63 Size preWindowSize() const { return Size(winWidth, winHeight); }
63 64 Size postWindowSize() const { return Size(winWidth + 1, winHeight + 1); }
... ... @@ -78,6 +79,20 @@ class MBLBPRepresentation : public Representation
78 79  
79 80 BR_REGISTER(Representation, MBLBPRepresentation)
80 81  
  82 +void MBLBPRepresentation::write(FileStorage &fs, const Mat &featureMap)
  83 +{
  84 + fs << "features" << "[";
  85 + const Mat_<int>& featureMap_ = (const Mat_<int>&)featureMap;
  86 + for ( int fi = 0; fi < featureMap.cols; fi++ )
  87 + if ( featureMap_(0, fi) >= 0 )
  88 + {
  89 + fs << "{";
  90 + features[fi].write( fs );
  91 + fs << "}";
  92 + }
  93 + fs << "]";
  94 +}
  95 +
81 96 MBLBPRepresentation::Feature::Feature( int offset, int x, int y, int _blockWidth, int _blockHeight )
82 97 {
83 98 Rect tr = rect = cvRect(x, y, _blockWidth, _blockHeight);
... ...