minor update of ar_hmdb code

modules/datasets/doc/datasets/ar_hmdb.rst

@@ -10,7 +10,7 @@ _`"HMDB: A Large Human Motion Database"`: http://serre-lab.clps.brown.edu/resour
 
  1. From link above download dataset files: hmdb51_org.rar & test_train_splits.rar.
 
- 2. Unpack them.
+ 2. Unpack them. Unpack all archives from directory: hmdb51_org/ and remove them.
 
  3. To load data run: ./opencv/build/bin/example_datasets_ar_hmdb -p=/home/user/path_to_unpacked_folders/
 
@@ -25,7 +25,7 @@ To run this benchmark execute:
 
  ./opencv/build/bin/example_datasets_ar_hmdb_benchmark -p=/home/user/path_to_unpacked_folders/
 
-(precomputed features should be unpacked in the same folder: /home/user/path_to_unpacked_folders/hmdb51_org_stips/)
+(precomputed features should be unpacked in the same folder: /home/user/path_to_unpacked_folders/hmdb51_org_stips/. Also unpack all archives from directory: hmdb51_org_stips/ and remove them.)
 
 **References:**
 

modules/datasets/include/opencv2/datasets/ar_hmdb.hpp

@@ -56,8 +56,9 @@ namespace datasets
 
 struct AR_hmdbObj : public Object
 {
+    int id;
     std::string name;
-    std::vector<std::string> videoNames;
+    std::string videoName;
 };
 
 class CV_EXPORTS AR_hmdb : public Dataset

modules/datasets/samples/ar_hmdb.cpp

@@ -74,15 +74,14 @@ int main(int argc, char *argv[])
     // And its size.
     int numSplits = dataset->getNumSplits();
     printf("splits number: %u\n", numSplits);
+    printf("train 1 size: %u\n", (unsigned int)dataset->getTrain(1).size());
+    printf("test 1 size: %u\n", (unsigned int)dataset->getTest(1).size());
 
     AR_hmdbObj *example = static_cast<AR_hmdbObj *>(dataset->getTrain(1)[0].get());
-    printf("name: %s\n", example->name.c_str());
-    vector<string> &videoNames = example->videoNames;
-    printf("size: %u\n", (unsigned int)videoNames.size());
-    for (vector<string>::iterator it=videoNames.begin(); it!=videoNames.end(); ++it)
-    {
-        printf("%s\n", (*it).c_str());
-    }
+    printf("first image:\n");
+    printf("action id: %u\n", example->id);
+    printf("action: %s\n", example->name.c_str());
+    printf("file: %s\n", example->videoName.c_str());
 
     return 0;
 }

modules/datasets/samples/ar_hmdb_benchmark.cpp

@@ -58,23 +58,6 @@ using namespace cv::datasets;
 using namespace cv::flann;
 using namespace cv::ml;
 
-unsigned int getNumFiles(vector< Ptr<Object> > &curr);
-unsigned int getNumFiles(vector< Ptr<Object> > &curr)
-{
-    unsigned int numFiles = 0;
-    for (unsigned int i=0; i<curr.size(); ++i)
-    {
-        AR_hmdbObj *example = static_cast<AR_hmdbObj *>(curr[i].get());
-        vector<string> &videoNames = example->videoNames;
-        for (vector<string>::iterator it=videoNames.begin(); it!=videoNames.end(); ++it)
-        {
-            numFiles++;
-        }
-    }
-
-    return numFiles;
-}
-
 void fillData(const string &path, vector< Ptr<Object> > &curr, Index &flann_index, Mat1f &data, Mat1i &labels);
 void fillData(const string &path, vector< Ptr<Object> > &curr, Index &flann_index, Mat1f &data, Mat1i &labels)
 {
@@ -87,36 +70,31 @@ void fillData(const string &path, vector< Ptr<Object> > &curr, Index &flann_inde
     for (unsigned int i=0; i<curr.size(); ++i)
     {
         AR_hmdbObj *example = static_cast<AR_hmdbObj *>(curr[i].get());
-        vector<string> &videoNames = example->videoNames;
-        for (vector<string>::iterator it=videoNames.begin(); it!=videoNames.end(); ++it)
+        string featuresFullPath = path + "hmdb51_org_stips/" + example->name + "/" + example->videoName + ".txt";
+
+        ifstream infile(featuresFullPath.c_str());
+        string line;
+        // skip header
+        for (unsigned int j=0; j<3; ++j)
         {
-            string featuresFile = *it + ".txt";
-            string featuresFullPath = path + "hmdb51_org_stips/" + example->name + "/" + featuresFile;
+            getline(infile, line);
+        }
+        while (getline(infile, line))
+        {
+            // 7 skip, hog+hof: 72+90 read
+            vector<string> elems;
+            split(line, elems, '\t');
 
-            ifstream infile(featuresFullPath.c_str());
-            string line;
-            // skip header
-            for (unsigned int j=0; j<3; ++j)
+            for (unsigned int j=0; j<descriptorNum; ++j)
             {
-                getline(infile, line);
+                sample(0, j) = (float)atof(elems[j+7].c_str());
             }
-            while (getline(infile, line))
-            {
-                // 7 skip, hog+hof: 72+90 read
-                vector<string> elems;
-                split(line, elems, '\t');
 
-                for (unsigned int j=0; j<descriptorNum; ++j)
-                {
-                    sample(0, j) = (float)atof(elems[j+7].c_str());
-                }
-
-                flann_index.knnSearch(sample, nresps, dists, 1, SearchParams());
-                data(numFiles, nresps(0, 0)) ++;
-            }
-            labels(numFiles, 0) = i;
-            numFiles++;
+            flann_index.knnSearch(sample, nresps, dists, 1, SearchParams());
+            data(numFiles, nresps(0, 0)) ++;
         }
+        labels(numFiles, 0) = example->id;
+        numFiles++;
     }
 }
 
@@ -148,43 +126,35 @@ int main(int argc, char *argv[])
     vector<double> res;
     for (int currSplit=0; currSplit<numSplits; ++currSplit)
     {
-
         Mat1f samples(sampleNum, descriptorNum);
         unsigned int currSample = 0;
         vector< Ptr<Object> > &curr = dataset->getTrain(currSplit);
-        unsigned int numTrainFiles = getNumFiles(curr);
         unsigned int numFeatures = 0;
         for (unsigned int i=0; i<curr.size(); ++i)
         {
             AR_hmdbObj *example = static_cast<AR_hmdbObj *>(curr[i].get());
-            vector<string> &videoNames = example->videoNames;
-            for (vector<string>::iterator it=videoNames.begin(); it!=videoNames.end(); ++it)
+            string featuresFullPath = path + "hmdb51_org_stips/" + example->name + "/" + example->videoName + ".txt";
+            ifstream infile(featuresFullPath.c_str());
+            string line;
+            // skip header
+            for (unsigned int j=0; j<3; ++j)
             {
-                string featuresFile = *it + ".txt";
-                string featuresFullPath = path + "hmdb51_org_stips/" + example->name + "/" + featuresFile;
-
-                ifstream infile(featuresFullPath.c_str());
-                string line;
-                // skip header
-                for (unsigned int j=0; j<3; ++j)
-                {
-                    getline(infile, line);
-                }
-                while (getline(infile, line))
+                getline(infile, line);
+            }
+            while (getline(infile, line))
+            {
+                numFeatures++;
+                if (currSample < sampleNum)
                 {
-                    numFeatures++;
-                    if (currSample < sampleNum)
-                    {
-                        // 7 skip, hog+hof: 72+90 read
-                        vector<string> elems;
-                        split(line, elems, '\t');
+                    // 7 skip, hog+hof: 72+90 read
+                    vector<string> elems;
+                    split(line, elems, '\t');
 
-                        for (unsigned int j=0; j<descriptorNum; ++j)
-                        {
-                            samples(currSample, j) = (float)atof(elems[j+7].c_str());
-                        }
-                        currSample++;
+                    for (unsigned int j=0; j<descriptorNum; ++j)
+                    {
+                        samples(currSample, j) = (float)atof(elems[j+7].c_str());
                     }
+                    currSample++;
                 }
             }
         }
@@ -202,6 +172,7 @@ int main(int argc, char *argv[])
         printf("resulted clusters number: %u\n", resultClusters);
 
 
+        unsigned int numTrainFiles = curr.size();
         Mat1f trainData(numTrainFiles, resultClusters);
         Mat1i trainLabels(numTrainFiles, 1);
 
@@ -232,7 +203,7 @@ int main(int argc, char *argv[])
 
         // prepare to predict
         curr = dataset->getTest(currSplit);
-        unsigned int numTestFiles = getNumFiles(curr);
+        unsigned int numTestFiles = curr.size();
         Mat1f testData(numTestFiles, resultClusters);
         Mat1i testLabels(numTestFiles, 1); // ground true
 
@@ -262,7 +233,6 @@ int main(int argc, char *argv[])
         double accuracy = 1.0*correct/numTestFiles;
         printf("correctly recognized actions: %f\n", accuracy);
         res.push_back(accuracy);
-
     }
 
     double accuracy = 0.0;

modules/datasets/src/ar_hmdb.cpp

@@ -42,6 +42,8 @@
 #include "opencv2/datasets/ar_hmdb.hpp"
 #include "opencv2/datasets/util.hpp"
 
+#include <map>
+
 namespace cv
 {
 namespace datasets
@@ -63,26 +65,9 @@ private:
 
     void loadDataset(const string &path);
 
-    void loadAction(const string &fileName, vector<string> &train_, vector<string> &test_);
+    map<string, int> actionsId;
 };
 
-void AR_hmdbImp::loadAction(const string &fileName, vector<string> &train_, vector<string> &test_)
-{
-    ifstream infile(fileName.c_str());
-    string video, label;
-    while (infile >> video >> label)
-    {
-        if ("1"==label)
-        {
-            train_.push_back(video);
-        } else
-        if ("2"==label)
-        {
-            test_.push_back(video);
-        }
-    }
-}
-
 /*AR_hmdbImp::AR_hmdbImp(const string &path, int number)
 {
     loadDataset(path, number);
@@ -120,18 +105,40 @@ void AR_hmdbImp::loadDatasetSplit(const string &path, int number)
     getDirList(pathDataset, fileNames);
     for (vector<string>::iterator it=fileNames.begin(); it!=fileNames.end(); ++it)
     {
-        Ptr<AR_hmdbObj> currTrain(new AR_hmdbObj);
-        Ptr<AR_hmdbObj> currTest(new AR_hmdbObj);
-        currTrain->name = *it;
-        currTest->name = *it;
-
-        train.back().push_back(currTrain);
-        test.back().push_back(currTest);
+        string &action = *it;
+        map<string, int>::iterator itId = actionsId.find(action);
+        int id;
+        if (itId == actionsId.end())
+        {
+            actionsId.insert(make_pair(action, actionsId.size()));
+            id = actionsId.size();
+        } else
+        {
+            id = (*itId).second;
+        }
 
         char tmp[2];
         sprintf(tmp, "%u", number+1);
-        string fileName(pathSplit + currTrain->name + "_test_split" + tmp + ".txt");
-        loadAction(fileName, currTrain->videoNames, currTest->videoNames);
+        string fileName(pathSplit + action + "_test_split" + tmp + ".txt");
+
+        ifstream infile(fileName.c_str());
+        string video, label;
+        while (infile >> video >> label)
+        {
+            Ptr<AR_hmdbObj> curr(new AR_hmdbObj);
+            curr->id = id;
+            curr->name = action;
+            curr->videoName = video;
+
+            if ("1"==label)
+            {
+                train.back().push_back(curr);
+            } else
+            if ("2"==label)
+            {
+                test.back().push_back(curr);
+            }
+        }
     }
 }