diff --git a/app/br/CMakeLists.txt b/app/br/CMakeLists.txt
index 128c671..398eb1d 100644
--- a/app/br/CMakeLists.txt
+++ b/app/br/CMakeLists.txt
@@ -4,6 +4,8 @@ endif()
 
 add_executable(br br.cpp ${BR_RESOURCES})
 target_link_libraries(br openbr ${CMAKE_THREAD_LIBS_INIT})
+qt5_use_modules(br ${QT_DEPENDENCIES})
+
 install(TARGETS br RUNTIME DESTINATION bin)
 
 add_test(NAME br_initialize WORKING_DIRECTORY ${CMAKE_BINARY_DIR} COMMAND br)
diff --git a/app/br/br.cpp b/app/br/br.cpp
index f16642c..c59ddb8 100644
--- a/app/br/br.cpp
+++ b/app/br/br.cpp
@@ -14,6 +14,9 @@
  * limitations under the License.                                            *
  * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
+#include <QCoreApplication>
+#include <QRunnable>
+#include <QThreadPool>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -47,171 +50,193 @@
  * \endcode
  */
 
-static void help()
+class FakeMain : public QRunnable
 {
-    printf("<arg> = Input; {arg} = Output; [arg] = Optional; (arg0|...|argN) = Choice\n"
-           "\n"
-           "==== Core Commands ====\n"
-           "-train <gallery> ... <gallery> [{model}]\n"
-           "-enroll <input_gallery> ... <input_gallery> {output_gallery}\n"
-           "-compare <target_gallery> <query_gallery> [{output}]\n"
-           "-eval <simmat> <mask> [{csv}]\n"
-           "-plot <file> ... <file> {destination}\n"
-           "\n"
-           "==== Other Commands ====\n"
-           "-fuse <simmat> ... <simmat> <mask> (None|MinMax|ZScore|WScore) (Min|Max|Sum[W1:W2:...:Wn]|Replace|Difference|None) {simmat}\n"
-           "-cluster <simmat> ... <simmat> <aggressiveness> {csv}\n"
-           "-makeMask <target_gallery> <query_gallery> {mask}\n"
-           "-combineMasks <mask> ... <mask> {mask} (And|Or)\n"
-           "-cat <gallery> ... <gallery> {gallery}\n"
-           "-convert <template> {template}\n"
-           "-reformat <target_sigset> <query_sigset> <simmat> {output}\n"
-           "-evalClassification <predicted_gallery> <truth_gallery>\n"
-           "-evalRegression <predicted_gallery> <truth_gallery>\n"
-           "-evalClusters <clusters> <sigset>\n"
-           "-confusion <file> <score>\n"
-           "-plotMetadata <file> ... <file> <columns>\n"
-           "\n"
-           "==== Configuration ====\n"
-           "-<key> <value>\n"
-           "\n"
-           "==== Miscellaneous ====\n"
-           "-objects [abstraction [implementation]]\n"
-           "-about\n"
-           "-version\n"
-           "-shell\n"
-           "-exit\n");
-}
+    int argc;
+    char **argv;
 
-static void check(bool condition, const char *error_message)
-{
-    if (!condition) {
-        printf("%s\n", error_message);
-        br_finalize();
-        exit(1);
-    }
-}
+public:
+    FakeMain(int argc_, char **argv_)
+        : argc(argc_), argv(argv_) {}
 
-int main(int argc, char *argv[])
-{
-    br_initialize(argc, argv);
+    void run()
+    {
+        // Remove program name
+        argv = &argv[1];
+        argc--;
 
-    // Remove program name
-    argv = &argv[1];
-    argc--;
-
-    if (argc == 0) printf("%s\nTry running 'br -help'\n", br_about());
-
-    bool shell = false;
-    while (shell || (argc > 0)) {
-        const char *fun;
-        int parc;
-        const char **parv;
-        if (shell) {
-            printf("> ");
-            br_read_line(&parc, &parv);
-            if (parc == 0) continue;
-            fun = parv[0];
-            parc--;
-            parv = &parv[1];
-        } else /* argc > 0 */ {
-            fun = &argv[0][1];
-            parc = 0; while ((1+parc < argc) && (argv[1+parc][0] != '-')) parc++;
-            parv = (const char **)&argv[1];
-            argc = argc - (1+parc);
-            argv = &argv[parc+1];
-        }
+        if (argc == 0) printf("%s\nTry running 'br -help'\n", br_about());
 
-        // Core Tasks
-        if (!strcmp(fun, "train")) {
-            check(parc >= 1, "Insufficient parameter count for 'train'.");
-            br_train_n(parc == 1 ? 1 : parc-1, parv, parc == 1 ? "" : parv[parc-1]);
-        } else if (!strcmp(fun, "enroll")) {
-            check(parc >= 1, "Insufficient parameter count for 'enroll'.");
-            if (parc == 1) br_enroll(parv[0]);
-            else             br_enroll_n(parc-1, parv, parv[parc-1]);
-        } else if (!strcmp(fun, "compare")) {
-            check((parc >= 2) && (parc <= 3), "Incorrect parameter count for 'compare'.");
-            br_compare(parv[0], parv[1], parc == 3 ? parv[2] : "");
-        } else if (!strcmp(fun, "eval")) {
-            check((parc >= 2) && (parc <= 3), "Incorrect parameter count for 'eval'.");
-            br_eval(parv[0], parv[1], parc == 3 ? parv[2] : "");
-        } else if (!strcmp(fun, "plot")) {
-            check(parc >= 2, "Incorrect parameter count for 'plot'.");
-            br_plot(parc-1, parv, parv[parc-1], true);
-        }
+        bool shell = false;
+        while (shell || (argc > 0)) {
+            const char *fun;
+            int parc;
+            const char **parv;
+            if (shell) {
+                printf("> ");
+                br_read_line(&parc, &parv);
+                if (parc == 0) continue;
+                fun = parv[0];
+                parc--;
+                parv = &parv[1];
+            } else /* argc > 0 */ {
+                fun = &argv[0][1];
+                parc = 0; while ((1+parc < argc) && (argv[1+parc][0] != '-')) parc++;
+                parv = (const char **)&argv[1];
+                argc = argc - (1+parc);
+                argv = &argv[parc+1];
+            }
+
+            // Core Tasks
+            if (!strcmp(fun, "train")) {
+                check(parc >= 1, "Insufficient parameter count for 'train'.");
+                br_train_n(parc == 1 ? 1 : parc-1, parv, parc == 1 ? "" : parv[parc-1]);
+            } else if (!strcmp(fun, "enroll")) {
+                check(parc >= 1, "Insufficient parameter count for 'enroll'.");
+                if (parc == 1) br_enroll(parv[0]);
+                else             br_enroll_n(parc-1, parv, parv[parc-1]);
+            } else if (!strcmp(fun, "compare")) {
+                check((parc >= 2) && (parc <= 3), "Incorrect parameter count for 'compare'.");
+                br_compare(parv[0], parv[1], parc == 3 ? parv[2] : "");
+            } else if (!strcmp(fun, "eval")) {
+                check((parc >= 2) && (parc <= 3), "Incorrect parameter count for 'eval'.");
+                br_eval(parv[0], parv[1], parc == 3 ? parv[2] : "");
+            } else if (!strcmp(fun, "plot")) {
+                check(parc >= 2, "Incorrect parameter count for 'plot'.");
+                br_plot(parc-1, parv, parv[parc-1], true);
+            }
 
-        // Secondary Tasks
-          else if (!strcmp(fun, "fuse")) {
-            check(parc >= 5, "Insufficient parameter count for 'fuse'.");
-            br_fuse(parc-4, parv, parv[parc-4], parv[parc-3], parv[parc-2], parv[parc-1]);
-        } else if (!strcmp(fun, "cluster")) {
-            check(parc >= 3, "Insufficient parameter count for 'cluster'.");
-            br_cluster(parc-2, parv, atof(parv[parc-2]), parv[parc-1]);
-        } else if (!strcmp(fun, "makeMask")) {
-            check(parc == 3, "Incorrect parameter count for 'makeMask'.");
-            br_make_mask(parv[0], parv[1], parv[2]);
-        } else if (!strcmp(fun, "combineMasks")) {
-            check(parc >= 4, "Insufficient parameter count for 'combineMasks'.");
-            br_combine_masks(parc-2, parv, parv[parc-2], parv[parc-1]);
-        } else if (!strcmp(fun, "cat")) {
-            check(parc >= 2, "Insufficient parameter count for 'cat'.");
-            br_cat(parc-1, parv, parv[parc-1]);
-        } else if (!strcmp(fun, "convert")) {
-            check(parc == 2, "Incorrect parameter count for 'convert'.");
-            br_convert(parv[0], parv[1]);
-        } else if (!strcmp(fun, "reformat")) {
-            check(parc == 4, "Incorrect parameter count for 'reformat'.");
-            br_reformat(parv[0], parv[1], parv[2], parv[3]);
-        } else if (!strcmp(fun, "evalClassification")) {
-            check(parc == 2, "Incorrect parameter count for 'evalClassification'.");
-            br_eval_classification(parv[0], parv[1]);
-        } else if (!strcmp(fun, "evalRegression")) {
-            check(parc == 2, "Incorrect parameter count for 'evalRegression'.");
-            br_eval_regression(parv[0], parv[1]);
-        } else if (!strcmp(fun, "evalClusters")) {
-            check(parc == 2, "Incorrect parameter count for 'evalClusters'.");
-            br_eval_clustering(parv[0], parv[1]);
-        } else if (!strcmp(fun, "confusion")) {
-            check(parc == 2, "Incorrect parameter count for 'confusion'.");
-            int true_positives, false_positives, true_negatives, false_negatives;
-            br_confusion(parv[0], atof(parv[1]),
-                         &true_positives, &false_positives, &true_negatives, &false_negatives);
-            printf("True Positives = %d\nFalse Positives = %d\nTrue Negatives = %d\nFalseNegatives = %d\n",
-                   true_positives, false_positives, true_negatives, false_negatives);
-        } else if (!strcmp(fun, "plotMetadata")) {
-            check(parc >= 2, "Incorrect parameter count for 'plotMetadata'.");
-            br_plot_metadata(parc-1, parv, parv[parc-1], true);
+            // Secondary Tasks
+            else if (!strcmp(fun, "fuse")) {
+                check(parc >= 5, "Insufficient parameter count for 'fuse'.");
+                br_fuse(parc-4, parv, parv[parc-4], parv[parc-3], parv[parc-2], parv[parc-1]);
+            } else if (!strcmp(fun, "cluster")) {
+                check(parc >= 3, "Insufficient parameter count for 'cluster'.");
+                br_cluster(parc-2, parv, atof(parv[parc-2]), parv[parc-1]);
+            } else if (!strcmp(fun, "makeMask")) {
+                check(parc == 3, "Incorrect parameter count for 'makeMask'.");
+                br_make_mask(parv[0], parv[1], parv[2]);
+            } else if (!strcmp(fun, "combineMasks")) {
+                check(parc >= 4, "Insufficient parameter count for 'combineMasks'.");
+                br_combine_masks(parc-2, parv, parv[parc-2], parv[parc-1]);
+            } else if (!strcmp(fun, "cat")) {
+                check(parc >= 2, "Insufficient parameter count for 'cat'.");
+                br_cat(parc-1, parv, parv[parc-1]);
+            } else if (!strcmp(fun, "convert")) {
+                check(parc == 2, "Incorrect parameter count for 'convert'.");
+                br_convert(parv[0], parv[1]);
+            } else if (!strcmp(fun, "reformat")) {
+                check(parc == 4, "Incorrect parameter count for 'reformat'.");
+                br_reformat(parv[0], parv[1], parv[2], parv[3]);
+            } else if (!strcmp(fun, "evalClassification")) {
+                check(parc == 2, "Incorrect parameter count for 'evalClassification'.");
+                br_eval_classification(parv[0], parv[1]);
+            } else if (!strcmp(fun, "evalRegression")) {
+                check(parc == 2, "Incorrect parameter count for 'evalRegression'.");
+                br_eval_regression(parv[0], parv[1]);
+            } else if (!strcmp(fun, "evalClusters")) {
+                check(parc == 2, "Incorrect parameter count for 'evalClusters'.");
+                br_eval_clustering(parv[0], parv[1]);
+            } else if (!strcmp(fun, "confusion")) {
+                check(parc == 2, "Incorrect parameter count for 'confusion'.");
+                int true_positives, false_positives, true_negatives, false_negatives;
+                br_confusion(parv[0], atof(parv[1]),
+                        &true_positives, &false_positives, &true_negatives, &false_negatives);
+                printf("True Positives = %d\nFalse Positives = %d\nTrue Negatives = %d\nFalseNegatives = %d\n",
+                       true_positives, false_positives, true_negatives, false_negatives);
+            } else if (!strcmp(fun, "plotMetadata")) {
+                check(parc >= 2, "Incorrect parameter count for 'plotMetadata'.");
+                br_plot_metadata(parc-1, parv, parv[parc-1], true);
+            }
+
+            // Miscellaneous
+            else if (!strcmp(fun, "help")) {
+                check(parc == 0, "No parameters expected for 'help'.");
+                help();
+            } else if (!strcmp(fun, "objects")) {
+                check(parc <= 2, "Incorrect parameter count for 'objects'.");
+                printf("%s\n", br_objects(parc >= 1 ? parv[0] : ".*", parc >= 2 ? parv[1] : ".*"));
+            } else if (!strcmp(fun, "about")) {
+                check(parc == 0, "No parameters expected for 'about'.");
+                printf("%s\n", br_about());
+            } else if (!strcmp(fun, "version")) {
+                check(parc == 0, "No parameters expected for 'version'.");
+                printf("%s\n", br_version());
+            } else if (!strcmp(fun, "shell")) {
+                check(parc == 0, "No parameters expected for 'shell'.");
+                shell = true;
+            } else if (!strcmp(fun, "exit")) {
+                check(parc == 0, "No parameters expected for 'exit'.");
+                shell = false;
+            } else if (!strcmp(fun, "br")) {
+                printf("That's me!\n");
+            } else if (parc <= 1) {
+                br_set_property(fun, parc >=1 ? parv[0] : "");
+            } else {
+                printf("Unrecognized function '%s'\n", fun);
+            }
         }
 
-        // Miscellaneous
-          else if (!strcmp(fun, "help")) {
-            check(parc == 0, "No parameters expected for 'help'.");
-            help();
-        } else if (!strcmp(fun, "objects")) {
-            check(parc <= 2, "Incorrect parameter count for 'objects'.");
-            printf("%s\n", br_objects(parc >= 1 ? parv[0] : ".*", parc >= 2 ? parv[1] : ".*"));
-        } else if (!strcmp(fun, "about")) {
-            check(parc == 0, "No parameters expected for 'about'.");
-            printf("%s\n", br_about());
-        } else if (!strcmp(fun, "version")) {
-            check(parc == 0, "No parameters expected for 'version'.");
-            printf("%s\n", br_version());
-        } else if (!strcmp(fun, "shell")) {
-            check(parc == 0, "No parameters expected for 'shell'.");
-            shell = true;
-        } else if (!strcmp(fun, "exit")) {
-            check(parc == 0, "No parameters expected for 'exit'.");
-            shell = false;
-        } else if (!strcmp(fun, "br")) {
-            printf("That's me!\n");
-        } else if (parc <= 1) {
-            br_set_property(fun, parc >=1 ? parv[0] : "");
-        } else {
-            printf("Unrecognized function '%s'\n", fun);
+        QCoreApplication::exit();
+    }
+
+private:
+    static void check(bool condition, const char *error_message)
+    {
+        if (!condition) {
+            printf("%s\n", error_message);
+            QCoreApplication::exit();
         }
     }
 
+    static void help()
+    {
+        printf("<arg> = Input; {arg} = Output; [arg] = Optional; (arg0|...|argN) = Choice\n"
+               "\n"
+               "==== Core Commands ====\n"
+               "-train <gallery> ... <gallery> [{model}]\n"
+               "-enroll <input_gallery> ... <input_gallery> {output_gallery}\n"
+               "-compare <target_gallery> <query_gallery> [{output}]\n"
+               "-eval <simmat> <mask> [{csv}]\n"
+               "-plot <file> ... <file> {destination}\n"
+               "\n"
+               "==== Other Commands ====\n"
+               "-fuse <simmat> ... <simmat> <mask> (None|MinMax|ZScore|WScore) (Min|Max|Sum[W1:W2:...:Wn]|Replace|Difference|None) {simmat}\n"
+               "-cluster <simmat> ... <simmat> <aggressiveness> {csv}\n"
+               "-makeMask <target_gallery> <query_gallery> {mask}\n"
+               "-combineMasks <mask> ... <mask> {mask} (And|Or)\n"
+               "-cat <gallery> ... <gallery> {gallery}\n"
+               "-convert <template> {template}\n"
+               "-reformat <target_sigset> <query_sigset> <simmat> {output}\n"
+               "-evalClassification <predicted_gallery> <truth_gallery>\n"
+               "-evalRegression <predicted_gallery> <truth_gallery>\n"
+               "-evalClusters <clusters> <sigset>\n"
+               "-confusion <file> <score>\n"
+               "-plotMetadata <file> ... <file> <columns>\n"
+               "\n"
+               "==== Configuration ====\n"
+               "-<key> <value>\n"
+               "\n"
+               "==== Miscellaneous ====\n"
+               "-objects [abstraction [implementation]]\n"
+               "-about\n"
+               "-version\n"
+               "-shell\n"
+               "-exit\n");
+    }
+};
+
+int main(int argc, char *argv[])
+{
+    br_initialize(argc, argv);
+
+    // Do argument execution in another thread so this main thread can run an event loop.
+    // When adding fakeMain to the global thread pool we give it a higher (slower) priority
+    //   so that it is preempted when parallel work is available,
+    //   otherwise we would only achieve (n-1)/n CPU utilization.
+    FakeMain *fakeMain = new FakeMain(argc, argv);
+    QThreadPool::globalInstance()->start(fakeMain, 1);
+    QCoreApplication::exec();
+
     br_finalize();
-    return 0;
 }
diff --git a/scripts/evalFaceRecognition-MEDS.sh b/scripts/evalFaceRecognition-MEDS.sh
index e339f92..f4dd134 100755
--- a/scripts/evalFaceRecognition-MEDS.sh
+++ b/scripts/evalFaceRecognition-MEDS.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 
-ALGORITHM=OpenBR
+ALGORITHM=FaceRecognitionHoG
 
 if [ ! -f evalFaceRecognition-MEDS.sh ]; then
   echo "Run this script from the scripts folder!"
diff --git a/scripts/trainAgeRegression-PCSO.sh b/scripts/trainAgeRegression-PCSO.sh
index 2a73e83..c063055 100755
--- a/scripts/trainAgeRegression-PCSO.sh
+++ b/scripts/trainAgeRegression-PCSO.sh
@@ -4,8 +4,8 @@ if [ ! -f trainAgeRegression-PCSO.sh ]; then
   exit
 fi
 
-#rm -f ../models/features/FaceClassificationRegistration
-#rm -f ../models/features/FaceClassificationExtraction
-rm -f ../models/algorithms/AgeRegression
+#rm -f ../share/openbr/models/features/FaceClassificationRegistration
+#rm -f ../share/openbr/models/features/FaceClassificationExtraction
+rm -f ../share/openbr/models/algorithms/AgeRegression
 
 br -algorithm AgeRegression -path ../data/PCSO/Images -train "../data/PCSO/PCSO.db[query='SELECT File,Age,PersonID FROM PCSO WHERE Age >= 15 AND AGE <= 75', subset=0:200]" ../share/openbr/models/algorithms/AgeRegression
diff --git a/scripts/trainFaceRecognition-PCSO.sh b/scripts/trainFaceRecognition-PCSO.sh
index fc09eb7..b46b127 100755
--- a/scripts/trainFaceRecognition-PCSO.sh
+++ b/scripts/trainFaceRecognition-PCSO.sh
@@ -4,10 +4,10 @@ if [ ! -f trainFaceRecognition-PCSO.sh ]; then
   exit
 fi
 
-#rm -f ../models/features/FaceRecognitionRegistration
-#rm -f ../models/features/FaceRecognitionExtraction
-#rm -f ../models/features/FaceRecognitionEmbedding
-#rm -f ../models/features/FaceRecognitionQuantization
-rm -f ../models/algorithms/FaceRecognition
+#rm -f ../share/openbr/models/features/FaceRecognitionRegistration
+#rm -f ../share/openbr/models/features/FaceRecognitionExtraction
+#rm -f ../share/openbr/models/features/FaceRecognitionEmbedding
+#rm -f ../share/openbr/models/features/FaceRecognitionQuantization
+rm -f ../share/openbr/models/algorithms/FaceRecognition
 
 br -algorithm FaceRecognition -path ../data/PCSO/Images -train "../data/PCSO/PCSO.db[query='SELECT File,'S'||PersonID,PersonID FROM PCSO', subset=0:5:6000]" ../share/openbr/models/algorithms/FaceRecognition
diff --git a/scripts/trainGenderClassification-PCSO.sh b/scripts/trainGenderClassification-PCSO.sh
index 88cfafe..91fae04 100755
--- a/scripts/trainGenderClassification-PCSO.sh
+++ b/scripts/trainGenderClassification-PCSO.sh
@@ -4,8 +4,8 @@ if [ ! -f trainGenderClassification-PCSO.sh ]; then
   exit
 fi
 
-#rm -f ../models/features/FaceClassificationRegistration
-#rm -f ../models/features/FaceClassificationExtraction
-rm -f ../models/algorithms/GenderClassification
+#rm -f ../share/openbr/models/features/FaceClassificationRegistration
+#rm -f ../share/openbr/models/features/FaceClassificationExtraction
+rm -f ../share/openbr/models/algorithms/GenderClassification
 
 br -algorithm GenderClassification -path ../data/PCSO/Images -train "../data/PCSO/PCSO.db[query='SELECT File,Gender,PersonID FROM PCSO', subset=0:8000]" ../share/openbr/models/algorithms/GenderClassification
diff --git a/scripts/trainImageRetrieval-LFW.sh b/scripts/trainImageRetrieval-LFW.sh
new file mode 100755
index 0000000..f0f11e5
--- /dev/null
+++ b/scripts/trainImageRetrieval-LFW.sh
@@ -0,0 +1,21 @@
+#!/bin/bash
+if [ ! -f trainImageRetrieval-LFW.sh ]; then
+  echo "Run this script from the scripts folder!"
+  exit
+fi
+
+if [ ! -e Algorithm_Dataset ]; then
+  mkdir Algorithm_Dataset
+fi
+
+if [ ! -e LFW.mask ]; then
+  br -makeMask ~/lfw2[step=10] . LFW.mask
+fi
+
+rm -f ../share/openbr/models/algorithms/ImageRetrieval
+
+br -algorithm ImageRetrieval -train ~/lfw2[step=10] ../share/openbr/models/algorithms/ImageRetrieval
+br -algorithm ImageRetrieval -compare ~/lfw2[step=10] . ImageRetrieval_LFW.mtx -eval ImageRetrieval_LFW.mtx LFW.mask Algorithm_Dataset/ImageRetrieval_LFW.csv
+br -plot Algorithm_Dataset/*_LFW.csv LFW
+
+#rm -f ../share/openbr/models/algorithms/ImageRetrieval
diff --git a/sdk/core/bee.cpp b/sdk/core/bee.cpp
index 982cdde..d6c5784 100644
--- a/sdk/core/bee.cpp
+++ b/sdk/core/bee.cpp
@@ -236,7 +236,7 @@ void BEE::makeMask(const QString &targetInput, const QString &queryInput, const 
     qDebug("Making mask from %s and %s to %s", qPrintable(targetInput), qPrintable(queryInput), qPrintable(mask));
 
     FileList targetFiles = TemplateList::fromGallery(targetInput).files();
-    FileList queryFiles = TemplateList::fromGallery(queryInput).files();
+    FileList queryFiles = (queryInput == ".") ? targetFiles : TemplateList::fromGallery(queryInput).files();
     QList<float> targetLabels = targetFiles.labels();
     QList<float> queryLabels = queryFiles.labels();
     QList<int> targetPartitions = targetFiles.crossValidationPartitions();
diff --git a/sdk/core/opencvutils.cpp b/sdk/core/opencvutils.cpp
index 9a178f3..c9a6364 100644
--- a/sdk/core/opencvutils.cpp
+++ b/sdk/core/opencvutils.cpp
@@ -143,6 +143,7 @@ Mat OpenCVUtils::toMatByRow(const QList<Mat> &src)
 
     int rows = 0; foreach (const Mat &m, src) rows += m.rows;
     int cols = src.first().cols;
+    if (cols == 0) qFatal("Columnless matrix!");
     int type = src.first().type();
     Mat dst(rows, cols, type);
 
@@ -172,22 +173,6 @@ QString OpenCVUtils::elemToString(const Mat &m, int r, int c)
     return "?";
 }
 
-float OpenCVUtils::elemToFloat(const Mat &m, int r, int c)
-{
-    assert(m.channels() == 1);
-    switch (m.depth()) {
-      case CV_8U:  return float(m.at<quint8>(r,c));
-      case CV_8S:  return float(m.at<qint8>(r,c));
-      case CV_16U: return float(m.at<quint16>(r,c));
-      case CV_16S: return float(m.at<qint16>(r,c));
-      case CV_32S: return float(m.at<qint32>(r,c));
-      case CV_32F: return float(m.at<float>(r,c));
-      case CV_64F: return float(m.at<double>(r,c));
-      default:     qFatal("Unknown matrix depth");
-    }
-    return 0;
-}
-
 QString OpenCVUtils::matrixToString(const Mat &m)
 {
     QString result;
@@ -224,18 +209,6 @@ QStringList OpenCVUtils::matrixToStringList(const Mat &m)
     return results;
 }
 
-QList<float> OpenCVUtils::matrixToVector(const Mat &m)
-{
-    QList<float> results;
-    vector<Mat> mv;
-    split(m, mv);
-    foreach (const Mat &mc, mv)
-        for (int i=0; i<mc.rows; i++)
-            for (int j=0; j<mc.cols; j++)
-                results.append(elemToFloat(mc, i, j));
-    return results;
-}
-
 Point2f OpenCVUtils::toPoint(const QPointF &qPoint)
 {
     return Point2f(qPoint.x(), qPoint.y());
diff --git a/sdk/core/opencvutils.h b/sdk/core/opencvutils.h
index 78ea238..72ac9c7 100644
--- a/sdk/core/opencvutils.h
+++ b/sdk/core/opencvutils.h
@@ -40,10 +40,38 @@ namespace OpenCVUtils
 
     // From image
     QString elemToString(const cv::Mat &m, int r, int c);
-    float elemToFloat(const cv::Mat &m, int r, int c);
     QString matrixToString(const cv::Mat &m);
     QStringList matrixToStringList(const cv::Mat &m);
-    QList<float> matrixToVector(const cv::Mat &m);
+
+    template <typename T>
+    T getElement(const cv::Mat &m, int r, int c)
+    {
+        assert(m.channels() == 1);
+        switch (m.depth()) {
+          case CV_8U:  return T(m.at<quint8>(r,c));
+          case CV_8S:  return T(m.at<qint8>(r,c));
+          case CV_16U: return T(m.at<quint16>(r,c));
+          case CV_16S: return T(m.at<qint16>(r,c));
+          case CV_32S: return T(m.at<qint32>(r,c));
+          case CV_32F: return T(m.at<float>(r,c));
+          case CV_64F: return T(m.at<double>(r,c));
+          default:     qFatal("Unknown matrix depth!");
+        }
+        return 0;
+    }
+
+    template <typename T>
+    QList<T> matrixToVector(const cv::Mat &m)
+    {
+        QList<T> results;
+        std::vector<cv::Mat> mv;
+        cv::split(m, mv);
+        foreach (const cv::Mat &mc, mv)
+            for (int i=0; i<mc.rows; i++)
+                for (int j=0; j<mc.cols; j++)
+                    results.append(getElement<float>(mc, i, j));
+        return results;
+    }
 
     // Conversions
     cv::Point2f toPoint(const QPointF &qPoint);
diff --git a/sdk/core/qtutils.cpp b/sdk/core/qtutils.cpp
index 8714e25..fa43afd 100644
--- a/sdk/core/qtutils.cpp
+++ b/sdk/core/qtutils.cpp
@@ -39,7 +39,7 @@ QStringList QtUtils::getFiles(QDir dir, bool recursive)
 
     QStringList files;
     foreach (const QString &file, NaturalStringSort(dir.entryList(QDir::Files)))
-        files.append(QDir::cleanPath(dir.absoluteFilePath(file)));
+        files.append(dir.absoluteFilePath(file));
 
     if (!recursive) return files;
 
diff --git a/sdk/core/qtutils.h b/sdk/core/qtutils.h
index 807ef43..96d6316 100644
--- a/sdk/core/qtutils.h
+++ b/sdk/core/qtutils.h
@@ -21,6 +21,7 @@
 #include <QDir>
 #include <QFile>
 #include <QFileInfo>
+#include <QFuture>
 #include <QMap>
 #include <QString>
 #include <QStringList>
diff --git a/sdk/frvt2012.cpp b/sdk/frvt2012.cpp
index 5be9f82..7ca43e7 100644
--- a/sdk/frvt2012.cpp
+++ b/sdk/frvt2012.cpp
@@ -29,7 +29,12 @@ static const int frvt2012_template_size = 768;
 
 static void initialize(const string &configuration_location)
 {
-    if (Globals == NULL) Context::initialize(0, NULL, QString::fromStdString(configuration_location));
+    // Fake the command line arguments
+    int argc = 1;
+    char arg1[1]; arg1[0]='\0';
+    char *argv[] = { arg1 };
+
+    if (Globals == NULL) Context::initialize(argc, argv, QString::fromStdString(configuration_location));
     Globals->quiet = true;
     Globals->parallelism = 0;
 }
diff --git a/sdk/openbr.cpp b/sdk/openbr.cpp
index a8897f9..c882399 100644
--- a/sdk/openbr.cpp
+++ b/sdk/openbr.cpp
@@ -103,7 +103,7 @@ void br_fuse(int num_input_simmats, const char *input_simmats[], const char *mas
     Fuse(QtUtils::toStringList(num_input_simmats, input_simmats), mask, normalization, fusion, output_simmat);
 }
 
-void br_initialize(int argc, char *argv[], const char *sdk_path)
+void br_initialize(int &argc, char *argv[], const char *sdk_path)
 {
     Context::initialize(argc, argv, sdk_path);
 }
diff --git a/sdk/openbr.h b/sdk/openbr.h
index c362328..5271098 100644
--- a/sdk/openbr.h
+++ b/sdk/openbr.h
@@ -211,7 +211,7 @@ BR_EXPORT void br_fuse(int num_input_simmats, const char *input_simmats[], const
  * \brief Wraps br::Context::initialize()
  * \see br_initialize_qt br_finalize
  */
-BR_EXPORT void br_initialize(int argc, char *argv[], const char *sdk_path = "");
+BR_EXPORT void br_initialize(int &argc, char *argv[], const char *sdk_path = "");
 
 /*!
  * \brief Wraps br::Context::initializeQt()
diff --git a/sdk/openbr_export.cpp b/sdk/openbr_export.cpp
index 13c9821..a3f8102 100644
--- a/sdk/openbr_export.cpp
+++ b/sdk/openbr_export.cpp
@@ -82,6 +82,7 @@ $ br -help
  *  -# Consider the free open source program <a href="http://wincdemu.sysprogs.org">WinCDEmu</a> if you need a program to mount ISO images.
  *  -# You will have to register with Microsoft after installation, but it's free.
  *  -# Grab any available <a href="http://www.microsoft.com/visualstudio/eng/downloads#d-visual-studio-2012-update">Visual Studio Updates</a>.
+ *  -# Download and install <a href="http://msdn.microsoft.com/en-us/windows/hardware/hh852363.aspx">Windows 8 SDK</a>.
  * -# <a href="http://www.cmake.org/files/v2.8/cmake-2.8.10.2-win32-x86.exe">Download CMake 2.8.10.2</a> and install.
  *  -# During installation setup select "add CMake to PATH".
  * -# <a href="http://downloads.sourceforge.net/project/opencvlibrary/opencv-unix/2.4.4/OpenCV-2.4.4.tar.bz2">Download OpenCV 2.4.4</a>.
@@ -100,19 +101,8 @@ $ br -help
  *  $ nmake install
  *  $ nmake clean
  *  \endcode
- * -# <a href="http://releases.qt-project.org/qt5/5.0.1/single/qt-everywhere-opensource-src-5.0.1.zip">Download Qt 5.0.1</a> and unzip.
- *  -# Install Perl/Python/Ruby dependencies as explained in the "Windows" section of "README". Make sure they are added to "path" when given the option during installation.
+ * -# <a href="http://releases.qt-project.org/digia/5.0.1/backups/2013-01-18-412/qt-windows-opensource-5.0.1-msvc2012_64-x86_64-offline-2013-01-18-412.exe">Download Qt 5.0.1</a> and install.
  *  -# <a href="http://www.microsoft.com/en-us/download/confirmation.aspx?id=6812">Download Direct X Software Developement Kit</a> and install.
- *  -# From the VS2012 x64 Cross Tools Command Prompt:
- *  \code
- *  $ cd qt-everywhere-opensource-src-5.0.1
- *  $ configure -prefix C:\Qt\5.0.1\msvc2012 -opensource -confirm-license -nomake examples -nomake tests
- *  $ nmake
- *  $ nmake install
- *  $ cd ..
- *  $ rmdir /Q /S qt-everywhere-opensource-src-5.0.1
- *  \endcode
- *  -# nmake will take several hours to finish.
  * -# Create a <a href="github.com">GitHub</a> account and follow their instructions for <a href="https://help.github.com/articles/set-up-git">setting up Git</a>.
  *  -# Launch "Git Bash" from the Desktop and clone OpenBR:
  *  \code
@@ -128,7 +118,7 @@ $ br -help
  *  $ cd C:\openbr
  *  $ mkdir build-msvc2012
  *  $ cd build-msvc2012
- *  $ cmake -G "CodeBlocks - NMake Makefiles" -DCMAKE_PREFIX_PATH="C:/OpenCV-2.4.4/build-msvc2012/install;C:/Qt/5.0.1/msvc2012" -DCMAKE_INSTALL_PREFIX="./install" -DBR_INSTALL_DEPENDENCIES=ON -DCMAKE_BUILD_TYPE=Release ..
+ *  $ cmake -G "CodeBlocks - NMake Makefiles" -DCMAKE_PREFIX_PATH="C:/OpenCV-2.4.4/build-msvc2012/install;C:/Qt/Qt5.0.1/5.0.1/msvc2012_64" -DCMAKE_INSTALL_PREFIX="./install" -DBR_INSTALL_DEPENDENCIES=ON -DCMAKE_BUILD_TYPE=Release ..
  *  $ nmake
  *  $ nmake install
  *  \endcode
@@ -147,7 +137,7 @@ $ br -help
  *  -# Browse to your prexisting build directory "C:\openbr\build-msvc2012" then select "Next".
  *  -# Clear any text in the "arguments" box then select "Run CMake" then "Finish".
  *  -# You're all set! You can find more information on Qt Creator <a href="http://qt-project.org/doc/qtcreator-2.6/">here</a> if you need.
- * -# Package OpenBR!
+ * -# (Optional) Package OpenBR!
  *  -# <a href="http://sourceforge.net/projects/nsis/files/NSIS%202/2.46/nsis-2.46-setup.exe/download?use_mirror=iweb&download=">Download NSIS 2.46</a> and install.
  *  -# From the VS2012 x64 Cross Tools Command Prompt:
  *  \code
@@ -226,7 +216,7 @@ $ br -help
  *  -# Browse to your prexisting build directory "C:\openbr\build-mingw64" then select "Next".
  *  -# Clear any text in the "arguments" box then select "Run CMake" then "Finish".
  *  -# You're all set! You can find more information on Qt Creator <a href="http://qt-project.org/doc/qtcreator-2.6/">here</a> if you need.
- * -# Package OpenBR!
+ * -# (Optional) Package OpenBR!
  *  -# <a href="http://sourceforge.net/projects/nsis/files/NSIS%202/2.46/nsis-2.46-setup.exe/download?use_mirror=iweb&download=">Download NSIS 2.46</a> and install.
  *  -# From the MinGW-w64 Command Prompt:
  *  \code
@@ -277,7 +267,7 @@ $ br -help
  * $ cd build
  * $ cmake -DCMAKE_PREFIX_PATH=~/Qt5.0.1/5.0.1/clang_64 -DCMAKE_BUILD_TYPE=Release ..
  * $ make -j4
- * $ make install
+ * $ sudo make install
  * \endcode
  * -# Hack OpenBR!
  *  -# Open Qt Creator IDE
@@ -289,12 +279,12 @@ $ br -help
  *  -# Browse to your prexisting build directory "~/openbr/build" then select "Continue".
  *  -# Select "Run CMake" then "Done".
  *  -# You're all set! You can find more information on Qt Creator <a href="http://qt-project.org/doc/qtcreator">here</a> if you need.
- * -# Package OpenBR!
+ * -# (Optional) Package OpenBR!
  * \code
  * $ cd openbr/build
  * $ make package
  * \endcode
- * -# Build OpenBR documentation!
+ * -# (Optional) Build OpenBR documentation!
  *  -# <a href="ftp://ftp.stack.nl/pub/users/dimitri/doxygen-1.8.2.src.tar.gz">Download Doxygen 1.8.2</a> and install:
  *  \code
  *  $ cd ~/Downloads
@@ -367,7 +357,7 @@ $ br -help
  * $ cd build
  * $ cmake -DCMAKE_PREFIX_PATH=~/Qt5.0.1/5.0.1/gcc_64 -DCMAKE_BUILD_TYPE=Release ..
  * $ make -j4
- * $ make install
+ * $ sudo make install
  * \endcode
  * -# Hack OpenBR!
  *  -# Open Qt Creator IDE
@@ -379,7 +369,7 @@ $ br -help
  *  -# Browse to your prexisting build directory "~/openbr/build" then select "Next".
  *  -# Select "Run CMake" then "Finish".
  *  -# You're all set! You can find more information on Qt Creator <a href="http://qt-project.org/doc/qtcreator-2.6/">here</a> if you need.
- * -# Package OpenBR!
+ * -# (Optional) Package OpenBR!
  * \code
  * $ cd openbr/build
  * $ make package
@@ -434,7 +424,7 @@ $ br -help
  * $ cd build-icc
  * $ cmake -DCMAKE_C_COMPILER=/opt/intel/bin/icc -DCMAKE_CXX_COMPILER=/opt/intel/bin/icpc -DCMAKE_PREFIX_PATH=~/Qt5.0.1/5.0.1/gcc_64 -DCMAKE_BUILD_TYPE=Release ..
  * $ make -j4
- * $ make install
+ * $ sudo make install
  * \endcode
  * -# Hack OpenBR!
  *  -# Open Qt Creator IDE
@@ -446,7 +436,7 @@ $ br -help
  *  -# Browse to your prexisting build directory "~/openbr/build" then select "Next".
  *  -# Select "Run CMake" then "Finish".
  *  -# You're all set! You can find more information on Qt Creator <a href="http://qt-project.org/doc/qtcreator-2.6/">here</a> if you need.
- * -# Package OpenBR!
+ * -# (Optional) Package OpenBR!
  * \code
  * $ cd openbr/build
  * $ make package
diff --git a/sdk/openbr_plugin.cpp b/sdk/openbr_plugin.cpp
index 612ad95..245bc12 100644
--- a/sdk/openbr_plugin.cpp
+++ b/sdk/openbr_plugin.cpp
@@ -14,6 +14,7 @@
  * limitations under the License.                                            *
  * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
+#include <QFutureSynchronizer>
 #include <QMetaProperty>
 #include <QPointF>
 #include <QRect>
@@ -28,6 +29,10 @@
 #include <iostream>
 #include <openbr_plugin.h>
 
+#ifndef BR_EMBEDDED
+#include <QApplication>
+#endif
+
 #include "version.h"
 #include "core/bee.h"
 #include "core/common.h"
@@ -385,6 +390,15 @@ TemplateList TemplateList::fromGallery(const br::File &gallery)
         QScopedPointer<Gallery> i(Gallery::make(file));
         TemplateList newTemplates = i->read();
         newTemplates = newTemplates.mid(gallery.get<int>("pos", 0), gallery.get<int>("length", -1));
+
+        const int step = gallery.get<int>("step", 1);
+        if (step > 1) {
+            TemplateList downsampled; downsampled.reserve(newTemplates.size()/step);
+            for (int i=0; i<newTemplates.size(); i+=step)
+                downsampled.append(newTemplates[i]);
+            newTemplates = downsampled;
+        }
+
         if (gallery.get<bool>("reduce", false)) newTemplates = newTemplates.reduced();
         const int crossValidate = gallery.get<int>("crossValidate");
         if (crossValidate > 0) srand(0);
@@ -770,27 +784,31 @@ int br::Context::timeRemaining() const
     return std::max(0, int((1 - p) / p * startTime.elapsed())) / 1000;
 }
 
-void br::Context::trackFutures(QList< QFuture<void> > &futures)
-{
-    foreach (QFuture<void> future, futures) {
-        QCoreApplication::processEvents();
-        future.waitForFinished();
-    }
-}
-
 bool br::Context::checkSDKPath(const QString &sdkPath)
 {
     return QFileInfo(sdkPath + "/share/openbr/openbr.bib").exists();
 }
 
-void br::Context::initialize(int argc, char *argv[], const QString &sdkPath)
+void br::Context::initialize(int &argc, char *argv[], const QString &sdkPath)
 {
+    // We take in argc as a reference due to:
+    //   https://bugreports.qt-project.org/browse/QTBUG-5637
+    // QApplication should be initialized before anything else.
+    // Since we can't ensure that it gets deleted last, we never delete it.
+    static QCoreApplication *application = NULL;
+    if (application == NULL) {
+#ifndef BR_EMBEDDED
+        application = new QApplication(argc, argv);
+#else
+        application = new QCoreApplication(argc, argv);
+#endif
+    }
+
     if (Globals == NULL) {
         Globals = new Context();
         Globals->init(File());
     }
 
-    Globals->coreApplication = QSharedPointer<QCoreApplication>(new QCoreApplication(argc, argv));
     initializeQt(sdkPath);
 
 #ifdef BR_DISTRIBUTED
@@ -853,7 +871,10 @@ void br::Context::initializeQt(QString sdkPath)
 
 void br::Context::finalize()
 {
-    // Trigger registerd finalizers
+    // Is anyone still running?
+    QThreadPool::globalInstance()->waitForDone();
+
+    // Trigger registered finalizers
     QList< QSharedPointer<Initializer> > initializers = Factory<Initializer>::makeAll();
     foreach (const QSharedPointer<Initializer> &initializer, initializers)
         initializer->finalize();
@@ -883,6 +904,11 @@ QString br::Context::scratchPath()
 
 void br::Context::messageHandler(QtMsgType type, const QMessageLogContext &context, const QString &msg)
 {
+    // Something about this method is not thread safe, and will lead to crashes if qDebug
+    // statements are called from multiple threads. Unless we lock the whole thing...
+    static QMutex generalLock;
+    QMutexLocker locker(&generalLock);
+
     QString txt;
     switch (type) {
       case QtDebugMsg:
@@ -904,11 +930,8 @@ void br::Context::messageHandler(QtMsgType type, const QMessageLogContext &conte
     Globals->mostRecentMessage = txt;
 
     if (Globals->logFile.isWritable()) {
-        static QMutex logLock;
-        logLock.lock();
         Globals->logFile.write(qPrintable(txt));
         Globals->logFile.flush();
-        logLock.unlock();
     }
 
     if (type == QtFatalMsg) {
@@ -917,8 +940,6 @@ void br::Context::messageHandler(QtMsgType type, const QMessageLogContext &conte
         Globals->finalize();
         abort();
     }
-
-    QCoreApplication::processEvents(); // Used to retrieve messages before event loop starts
 }
 
 Context *br::Globals = NULL;
@@ -1147,14 +1168,12 @@ private:
         for (int i=0; i<templatesList.size(); i++)
             templatesList[i] = Downsample(templatesList[i], transforms[i]);
 
-        QList< QFuture<void> > futures;
-        const bool threaded = Globals->parallelism && (templatesList.size() > 1);
+        QFutureSynchronizer<void> futures;
         for (int i=0; i<templatesList.size(); i++) {
-            if (threaded) futures.append(QtConcurrent::run(_train, transforms[i], &templatesList[i]));
-            else                                           _train (transforms[i], &templatesList[i]);
+            if (Globals->parallelism) futures.addFuture(QtConcurrent::run(_train, transforms[i], &templatesList[i]));
+            else                                                          _train (transforms[i], &templatesList[i]);
         }
-
-        if (threaded) Globals->trackFutures(futures);
+        futures.waitForFinished();
     }
 
     void project(const Template &src, Template &dst) const
@@ -1250,42 +1269,51 @@ Transform *Transform::clone() const
     return clone;
 }
 
-static void _backProject(const Transform *transform, const Template *dst, Template *src)
+static void _project(const Transform *transform, const Template *src, Template *dst)
 {
     try {
-        transform->backProject(*dst, *src);
+        transform->project(*src, *dst);
     } catch (...) {
         qWarning("Exception triggered when processing %s with transform %s", qPrintable(src->file.flat()), qPrintable(transform->objectName()));
-        *src = Template(dst->file);
-        src->file.set("FTE", true);
+        *dst = Template(src->file);
+        dst->file.set("FTE", true);
     }
 }
 
-
-
-// Default project(TemplateList) -- call project(template) separately for each input
-// template
+// Default project(TemplateList) calls project(Template) separately for each element
 void Transform::project(const TemplateList &src, TemplateList &dst) const
 {
-    dst.clear();
-
-    // Project templates derived from a single image, default implementation: project each
-    // input template to an ouptut template individually.
-    foreach(const Template  & src_template, src) {
-        dst.append(Template(src_template.file));
-        try {
-                project(src_template, dst.back());
-        } catch (...) {
-            qWarning("Exception triggered when processing %s with transform %s", qPrintable(src_template.file.flat()), qPrintable(objectName()));
-            dst.back() = Template(src_template.file);
-            dst.back().file.set("FTE", true);
+    dst.reserve(src.size());
+
+    // There are certain conditions where we should process the templates in serial,
+    // but generally we'd prefer to process them in parallel.
+    if ((src.size() < 2) ||
+        (QThreadPool::globalInstance()->activeThreadCount() >= QThreadPool::globalInstance()->maxThreadCount()) ||
+        (Globals->parallelism == 0)) {
+
+        foreach (const Template &t, src) {
+            dst.append(Template());
+            _project(this, &t, &dst.last());
         }
+    } else {
+        for (int i=0; i<src.size(); i++)
+            dst.append(Template());
+        QFutureSynchronizer<void> futures;
+        for (int i=0; i<dst.size(); i++)
+            futures.addFuture(QtConcurrent::run(_project, this, &src[i], &dst[i]));
+        futures.waitForFinished();
     }
 }
 
-void Transform::backProject(const Template &dst, Template &src) const
+static void _backProject(const Transform *transform, const Template *dst, Template *src)
 {
-    src = dst;
+    try {
+        transform->backProject(*dst, *src);
+    } catch (...) {
+        qWarning("Exception triggered when processing %s with transform %s", qPrintable(src->file.flat()), qPrintable(transform->objectName()));
+        *src = Template(dst->file);
+        src->file.set("FTE", true);
+    }
 }
 
 void Transform::backProject(const TemplateList &dst, TemplateList &src) const
@@ -1293,12 +1321,11 @@ void Transform::backProject(const TemplateList &dst, TemplateList &src) const
     src.reserve(dst.size());
     for (int i=0; i<dst.size(); i++) src.append(Template());
 
-    QList< QFuture<void> > futures;
-    if (Globals->parallelism) futures.reserve(dst.size());
+    QFutureSynchronizer<void> futures;
     for (int i=0; i<dst.size(); i++)
-        if (Globals->parallelism) futures.append(QtConcurrent::run(_backProject, this, &dst[i], &src[i]));
-        else                                                      _backProject (this, &dst[i], &src[i]);
-    if (Globals->parallelism) Globals->trackFutures(futures);
+        if (Globals->parallelism) futures.addFuture(QtConcurrent::run(_backProject, this, &dst[i], &src[i]));
+        else                                                          _backProject (this, &dst[i], &src[i]);
+    futures.waitForFinished();
 }
 
 /* Distance - public methods */
@@ -1326,16 +1353,16 @@ void Distance::compare(const TemplateList &target, const TemplateList &query, Ou
     const bool stepTarget = target.size() > query.size();
     const int totalSize = std::max(target.size(), query.size());
     int stepSize = ceil(float(totalSize) / float(std::max(1, abs(Globals->parallelism))));
-    QList< QFuture<void> > futures; futures.reserve(ceil(float(totalSize)/float(stepSize)));
+    QFutureSynchronizer<void> futures;
     for (int i=0; i<totalSize; i+=stepSize) {
         const TemplateList &targets(stepTarget ? TemplateList(target.mid(i, stepSize)) : target);
         const TemplateList &queries(stepTarget ? query : TemplateList(query.mid(i, stepSize)));
         const int targetOffset = stepTarget ? i : 0;
         const int queryOffset = stepTarget ? 0 : i;
-        if (Globals->parallelism) futures.append(QtConcurrent::run(this, &Distance::compareBlock, targets, queries, output, targetOffset, queryOffset));
-        else                                                                        compareBlock (targets, queries, output, targetOffset, queryOffset);
+        if (Globals->parallelism) futures.addFuture(QtConcurrent::run(this, &Distance::compareBlock, targets, queries, output, targetOffset, queryOffset));
+        else                                                                           compareBlock (targets, queries, output, targetOffset, queryOffset);
     }
-    if (Globals->parallelism) Globals->trackFutures(futures);
+    futures.waitForFinished();
 }
 
 QList<float> Distance::compare(const TemplateList &targets, const Template &query) const
diff --git a/sdk/openbr_plugin.h b/sdk/openbr_plugin.h
index 89cdd24..4e63952 100644
--- a/sdk/openbr_plugin.h
+++ b/sdk/openbr_plugin.h
@@ -17,7 +17,6 @@
 #ifndef __OPENBR_PLUGIN_H
 #define __OPENBR_PLUGIN_H
 
-#include <QCoreApplication>
 #include <QDataStream>
 #include <QDebug>
 #include <QDir>
@@ -466,7 +465,6 @@ private:
 class BR_EXPORT Context : public Object
 {
     Q_OBJECT
-    QSharedPointer<QCoreApplication> coreApplication;
     QFile logFile;
 
 public:
@@ -624,12 +622,6 @@ public:
     int timeRemaining() const;
 
     /*!
-     * \brief Continues to print the progress of the futures until they are completed.
-     * \param futures The list of futures to track.
-     */
-    void trackFutures(QList< QFuture<void> > &futures);
-
-    /*!
      * \brief Returns \c true if \em sdkPath is valid, \c false otherwise.
      * \param sdkPath The path to <tt>share/openbr/openbr.bib</tt>
      */
@@ -658,7 +650,7 @@ public:
      * \note <a href="http://qt-project.org/">Qt</a> users should instead call initializeQt().
      * \see initializeQt finalize
      */
-    static void initialize(int argc, char *argv[], const QString &sdkPath = "");
+    static void initialize(int &argc, char *argv[], const QString &sdkPath = "");
 
     /*!
      * \brief Alternative to initialize() for <a href="http://qt-project.org/">Qt</a> users.
@@ -998,7 +990,7 @@ public:
     virtual void train(const TemplateList &data) = 0; /*!< \brief Train the transform. */
     virtual void project(const Template &src, Template &dst) const = 0; /*!< \brief Apply the transform. */
     virtual void project(const TemplateList &src, TemplateList &dst) const; /*!< \brief Apply the transform. */
-    virtual void backProject(const Template &dst, Template &src) const; /*!< \brief Invert the transform. */
+    virtual void backProject(const Template &dst, Template &src) const { src = dst; } /*!< \brief Invert the transform. */
     virtual void backProject(const TemplateList &dst, TemplateList &src) const; /*!< \brief Invert the transform. */
 
     /*!< \brief Apply the transform, may update the transform's internal state */
@@ -1113,6 +1105,7 @@ class BR_EXPORT TimeVaryingTransform : public Transform
 {
     Q_OBJECT
 
+public:
     virtual bool timeVarying() const { return true; }
 
     virtual void project(const Template &src, Template &dst) const
@@ -1127,6 +1120,23 @@ class BR_EXPORT TimeVaryingTransform : public Transform
         (void) dst; (void) src;
     }
 
+    // Get a compile failure if this isn't here to go along with the other
+    // projectUpdate, no idea why
+    virtual void projectUpdate(const Template & src, Template & dst)
+    {
+        (void) src; (void) dst;
+        qFatal("do something useful");
+    }
+
+    virtual void projectUpdate(const TemplateList &src, TemplateList &dst)
+    {
+        foreach (const Template & src_part, src) {
+            Template out;
+            projectUpdate(src_part, out);
+            dst.append(out);
+        }
+    }
+
 protected:
     TimeVaryingTransform(bool independent = true, bool trainable = true) : Transform(independent, trainable) {}
 };
@@ -1170,6 +1180,51 @@ protected:
     UntrainableMetaTransform() : UntrainableTransform(false) {}
 };
 
+/*!
+ * \brief A MetaTransform that aggregates some sub-transforms
+ */
+class BR_EXPORT CompositeTransform : public TimeVaryingTransform
+{
+    Q_OBJECT
+
+public:
+    Q_PROPERTY(QList<br::Transform*> transforms READ get_transforms WRITE set_transforms RESET reset_transforms)
+    BR_PROPERTY(QList<br::Transform*>, transforms, QList<br::Transform*>())
+
+    virtual void project(const Template &src, Template &dst) const
+    {
+        if (timeVarying()) qFatal("No const project defined for time-varying transform");
+        _project(src, dst);
+    }
+
+    virtual void project(const TemplateList &src, TemplateList &dst) const
+    {
+        if (timeVarying()) qFatal("No const project defined for time-varying transform");
+        _project(src, dst);
+    }
+
+    bool timeVarying() const { return isTimeVarying; }
+
+    void init()
+    {
+        isTimeVarying = false;
+        foreach (const br::Transform *transform, transforms) {
+            if (transform->timeVarying()) {
+                isTimeVarying = true;
+                break;
+            }
+        }
+    }
+
+protected:
+    bool isTimeVarying;
+
+    virtual void _project(const Template & src, Template & dst) const = 0;
+    virtual void _project(const TemplateList & src, TemplateList & dst) const = 0;
+
+    CompositeTransform() : TimeVaryingTransform(false) {}
+};
+
 /*! @}*/
 
 /*!
diff --git a/sdk/plugins/algorithms.cpp b/sdk/plugins/algorithms.cpp
index 526bcb7..80a9a8c 100644
--- a/sdk/plugins/algorithms.cpp
+++ b/sdk/plugins/algorithms.cpp
@@ -32,7 +32,6 @@ class AlgorithmsInitializer : public Initializer
     {
         // Face
         Globals->abbreviations.insert("FaceRecognition", "FaceDetection!<FaceRecognitionRegistration>!<FaceRecognitionExtraction>+<FaceRecognitionEmbedding>+<FaceRecognitionQuantization>:MatchProbability(ByteL1)");
-        Globals->abbreviations.insert("FaceRecognitionNoTraining", "FaceDetection!ASEFEyes+Affine(86,86,0.25,0.35)!Blur(1.1)+Gamma(0.2)+DoG(1,2)+ContrastEq(0.1,10)+Mask+LBP(1,2)+RectRegions(8,8,6,6)+Hist(59)+Cat:ChiSquared");
         Globals->abbreviations.insert("GenderClassification", "FaceDetection!<FaceClassificationRegistration>!<FaceClassificationExtraction>+<GenderClassifier>+Discard");
         Globals->abbreviations.insert("AgeRegression", "FaceDetection!<FaceClassificationRegistration>!<FaceClassificationExtraction>+<AgeRegressor>+Discard");
         Globals->abbreviations.insert("FaceQuality", "Open!Cascade(FrontalFace)+ASEFEyes+Affine(64,64,0.25,0.35)+ImageQuality+Cvt(Gray)+DFFS+Discard");
@@ -43,6 +42,7 @@ class AlgorithmsInitializer : public Initializer
         Globals->abbreviations.insert("OpenBR", "FaceRecognition");
         Globals->abbreviations.insert("GenderEstimation", "GenderClassification");
         Globals->abbreviations.insert("AgeEstimation", "AgeRegression");
+        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)");
 
         // Generic Image Processing
         Globals->abbreviations.insert("SIFT", "Open+KeyPointDetector(SIFT)+KeyPointDescriptor(SIFT):KeyPointMatcher(BruteForce)");
@@ -50,7 +50,7 @@ class AlgorithmsInitializer : public Initializer
         Globals->abbreviations.insert("SmallSIFT", "Open+LimitSize(512)+KeyPointDetector(SIFT)+KeyPointDescriptor(SIFT):KeyPointMatcher(BruteForce)");
         Globals->abbreviations.insert("SmallSURF", "Open+LimitSize(512)+KeyPointDetector(SURF)+KeyPointDescriptor(SURF):KeyPointMatcher(BruteForce)");
         Globals->abbreviations.insert("ColorHist", "Open+LimitSize(512)!EnsureChannels(3)+SplitChannels+Hist(256,0,8)+Cat+Normalize(L1):L2");
-        Globals->abbreviations.insert("IHH", "Open+(RG+MAdd(0.5))/(Cvt(Gray)+Gradient+Bin(0,360,8,true))+Merge+Integral+IntegralSampler+CvtFloat+WordWise(RowWisePCA(8)+RowWiseMeanCenter+Binarize,Identity):L2");
+        Globals->abbreviations.insert("ImageRetrieval", "Open+Cvt(Gray)+Cascade(FrontalFace)+ASEFEyes+Affine(88,88,0.25,0.35)+Gradient+Bin(0,360,8,true)+Merge+Integral+IntegralSampler+WordWise(RowWisePCA(8)+RowWiseMeanCenter+Binarize,RowWisePCA)+Sentence:SentenceSimilarity");
 
         // Hash
         Globals->abbreviations.insert("FileName", "Name+Identity:Identical");
diff --git a/sdk/plugins/cvt.cpp b/sdk/plugins/cvt.cpp
index b48ed56..52314fa 100644
--- a/sdk/plugins/cvt.cpp
+++ b/sdk/plugins/cvt.cpp
@@ -181,12 +181,17 @@ class RGTransform : public UntrainableTransform
         for (int i=0; i<m.rows; i++)
             for (int j=0; j<m.cols; j++) {
                 Vec3b v = m.at<Vec3b>(i,j);
-                uchar& b = v[0];
-                uchar& g = v[1];
-                uchar& r = v[2];
-
-                R.at<uchar>(i, j) = saturate_cast<uchar>(255.0*r/(r+g+b));
-                G.at<uchar>(i, j) = saturate_cast<uchar>(255.0*g/(r+g+b));
+                const int b = v[0];
+                const int g = v[1];
+                const int r = v[2];
+                const int sum = b + g + r;
+                if (sum > 0) {
+                    R.at<uchar>(i, j) = saturate_cast<uchar>(255.0*r/(r+g+b));
+                    G.at<uchar>(i, j) = saturate_cast<uchar>(255.0*g/(r+g+b));
+                } else {
+                    R.at<uchar>(i, j) = 0;
+                    G.at<uchar>(i, j) = 0;
+                }
             }
 
         dst.append(R);
diff --git a/sdk/plugins/distance.cpp b/sdk/plugins/distance.cpp
index 734dd95..5df5bdc 100644
--- a/sdk/plugins/distance.cpp
+++ b/sdk/plugins/distance.cpp
@@ -14,11 +14,13 @@
  * limitations under the License.                                            *
  * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
+#include <QFutureSynchronizer>
 #include <QtConcurrentRun>
 #include <opencv2/imgproc/imgproc.hpp>
 #include <openbr_plugin.h>
 
 #include "core/distance_sse.h"
+#include "core/qtutils.h"
 
 using namespace cv;
 
@@ -154,11 +156,11 @@ class PipeDistance : public Distance
 
     void train(const TemplateList &data)
     {
-        QList< QFuture<void> > futures;
+        QFutureSynchronizer<void> futures;
         foreach (br::Distance *distance, distances)
-            if (Globals->parallelism) futures.append(QtConcurrent::run(distance, &Distance::train, data));
-            else                      distance->train(data);
-        Globals->trackFutures(futures);
+            if (Globals->parallelism) futures.addFuture(QtConcurrent::run(distance, &Distance::train, data));
+            else                                                          distance->train(data);
+        futures.waitForFinished();
     }
 
     float compare(const Template &a, const Template &b) const
diff --git a/sdk/plugins/eigen3.cpp b/sdk/plugins/eigen3.cpp
index 1368d25..8aede98 100644
--- a/sdk/plugins/eigen3.cpp
+++ b/sdk/plugins/eigen3.cpp
@@ -145,15 +145,15 @@ protected:
 
         if (keep < 1) {
             // Keep eigenvectors that retain a certain energy percentage.
-            double totalEnergy = allEVals.sum();
+            const double totalEnergy = allEVals.sum();
             if (totalEnergy == 0) {
                 keep = 0;
             } else {
                 double currentEnergy = 0;
-                int i;
-                for (i=1; i<=allEVals.rows(); i++) {
-                    currentEnergy += allEVals(allEVals.rows()-i);
-                    if (currentEnergy / totalEnergy >= keep) break;
+                int i=0;
+                while ((currentEnergy / totalEnergy < keep) && (i < allEVals.rows())) {
+                    currentEnergy += allEVals(allEVals.rows()-(i+1));
+                    i++;
                 }
                 keep = i - drop;
             }
diff --git a/sdk/plugins/format.cpp b/sdk/plugins/format.cpp
index 680c452..c460746 100644
--- a/sdk/plugins/format.cpp
+++ b/sdk/plugins/format.cpp
@@ -242,12 +242,11 @@ class DefaultFormat : public Format
 
     void write(const Template &t) const
     {
-        if (t.size() != 1) {
+        if (t.size() > 1) {
             videoFormat videoWriter;
             videoWriter.file = file;
             videoWriter.write(t);
-        }
-        else {
+        } else if (t.size() == 1) {
             imwrite(file.name.toStdString(), t);
         }
     }
@@ -565,7 +564,6 @@ class webcamFormat : public Format
 
 BR_REGISTER(Format, webcamFormat)
 
-#ifndef BR_EMBEDDED
 /*!
  * \ingroup formats
  * \brief Decodes images from Base64 xml
@@ -578,13 +576,15 @@ class xmlFormat : public Format
 
     Template read() const
     {
+        Template t;
+
+#ifndef BR_EMBEDDED
         QDomDocument doc(file);
         QFile f(file);
         if (!f.open(QIODevice::ReadOnly)) qFatal("Unable to open %s for reading.", qPrintable(file.flat()));
         if (!doc.setContent(&f))          qFatal("Unable to parse %s.", qPrintable(file.flat()));
         f.close();
 
-        Template t;
         QDomElement docElem = doc.documentElement();
         QDomNode subject = docElem.firstChild();
         while (!subject.isNull()) {
@@ -620,6 +620,7 @@ class xmlFormat : public Format
             if (current.month() < dob.month()) age--;
             t.file.set("Age", age);
         }
+#endif // BR_EMBEDDED
 
         return t;
     }
@@ -632,7 +633,6 @@ class xmlFormat : public Format
 };
 
 BR_REGISTER(Format, xmlFormat)
-#endif // BR_EMBEDDED
 
 } // namespace br
 
diff --git a/sdk/plugins/gallery.cpp b/sdk/plugins/gallery.cpp
index 03f2c42..e522d44 100644
--- a/sdk/plugins/gallery.cpp
+++ b/sdk/plugins/gallery.cpp
@@ -14,6 +14,7 @@
  * limitations under the License.                                            *
  * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
+#include <QtConcurrentRun>
 #ifndef BR_EMBEDDED
 #include <QNetworkAccessManager>
 #include <QNetworkReply>
@@ -103,9 +104,14 @@ class EmptyGallery : public Gallery
 
         // Add immediate subfolders
         QDir dir(file);
-        foreach (const QString &folder, NaturalStringSort(dir.entryList(QDir::Dirs | QDir::NoDotAndDotDot)))
-            foreach (const QString &file, QtUtils::getFiles(dir.absoluteFilePath(folder), true))
-                templates.append(File(file, folder));
+        QList< QFuture<TemplateList> > futures;
+        foreach (const QString &folder, NaturalStringSort(dir.entryList(QDir::Dirs | QDir::NoDotAndDotDot))) {
+            const QDir subdir = dir.absoluteFilePath(folder);
+            if (Globals->parallelism) futures.append(QtConcurrent::run(&EmptyGallery::getTemplates, subdir));
+            else                      templates.append(getTemplates(subdir));
+        }
+        foreach (const QFuture<TemplateList> &future, futures)
+            templates.append(future.result());
 
         // Add root folder
         foreach (const QString &fileName, QtUtils::getFiles(file.name, false))
@@ -130,6 +136,15 @@ class EmptyGallery : public Gallery
             format->write(t);
         }
     }
+
+    static TemplateList getTemplates(const QDir &dir)
+    {
+        const QStringList files = QtUtils::getFiles(dir, true);
+        TemplateList templates; templates.reserve(files.size());
+        foreach (const QString &file, files)
+            templates.append(File(file, dir.dirName()));
+        return templates;
+    }
 };
 
 BR_REGISTER(Gallery, EmptyGallery)
@@ -662,7 +677,7 @@ class googleGallery : public Gallery
             QNetworkReply *reply = networkAccessManager.get(request);
 
             while (!reply->isFinished())
-                QCoreApplication::processEvents();
+                QThread::yieldCurrentThread();
 
             QString data(reply->readAll());
             delete reply;
diff --git a/sdk/plugins/gui.cmake b/sdk/plugins/gui.cmake
new file mode 100644
index 0000000..af09d1b
--- /dev/null
+++ b/sdk/plugins/gui.cmake
@@ -0,0 +1,3 @@
+if(NOT ${BR_EMBEDDED})
+  set(BR_THIRDPARTY_SRC ${BR_THIRDPARTY_SRC} plugins/gui.cpp)
+endif()
diff --git a/sdk/plugins/gui.cpp b/sdk/plugins/gui.cpp
new file mode 100644
index 0000000..13dc723
--- /dev/null
+++ b/sdk/plugins/gui.cpp
@@ -0,0 +1,159 @@
+#include <QApplication>
+#include <QLabel>
+#include <opencv2/imgproc/imgproc.hpp>
+#include <openbr_plugin.h>
+
+using namespace cv;
+
+namespace br
+{
+QImage toQImage(const Mat &mat)
+{
+    // Convert to 8U depth
+    Mat mat8u;
+    if (mat.depth() != CV_8U) {
+        double globalMin = std::numeric_limits<double>::max();
+        double globalMax = -std::numeric_limits<double>::max();
+
+        std::vector<Mat> mv;
+        split(mat, mv);
+        for (size_t i=0; i<mv.size(); i++) {
+            double min, max;
+            minMaxLoc(mv[i], &min, &max);
+            globalMin = std::min(globalMin, min);
+            globalMax = std::max(globalMax, max);
+        }
+        assert(globalMax >= globalMin);
+
+        double range = globalMax - globalMin;
+        if (range != 0) {
+            double scale = 255 / range;
+            convertScaleAbs(mat, mat8u, scale, -(globalMin * scale));
+        } else {
+            // Monochromatic
+            mat8u = Mat(mat.size(), CV_8UC1, Scalar((globalMin+globalMax)/2));
+        }
+    } else {
+        mat8u = mat;
+    }
+
+    // Convert to 3 channels
+    Mat mat8uc3;
+    if      (mat8u.channels() == 4) cvtColor(mat8u, mat8uc3, CV_BGRA2RGB);
+    else if (mat8u.channels() == 3) cvtColor(mat8u, mat8uc3, CV_BGR2RGB);
+    else if (mat8u.channels() == 1) cvtColor(mat8u, mat8uc3, CV_GRAY2RGB);
+
+    return QImage(mat8uc3.data, mat8uc3.cols, mat8uc3.rows, 3*mat8uc3.cols, QImage::Format_RGB888).copy();
+}
+
+// Provides slots for manipulating a QLabel, but does not inherit from QWidget.
+// Therefore, it can be moved to the main thread if not created there initially
+// since god forbid you create a QWidget subclass in not the main thread.
+class GUIProxy : public QObject
+{
+    Q_OBJECT
+public:
+    QLabel * window;
+
+    GUIProxy()
+    {
+        window = NULL;
+    }
+
+public slots:
+
+    void showImage(const QPixmap & input)
+    {
+        window->setPixmap(input);
+        window->setFixedSize(input.size());
+        window->setVisible(true);
+    }
+
+    void createWindow()
+    {
+        delete window;
+        window = new QLabel();
+    }
+};
+
+/*!
+ * \ingroup transforms
+ * \brief Displays templates in a GUI pop-up window using QT.
+ * \author Charles Otto \cite caotto
+ * Unlike ShowTransform, this can be used with parallelism enabled, although it
+ * is considered TimeVarying.
+ */
+class Show2Transform : public TimeVaryingTransform
+{
+    Q_OBJECT
+public:
+    Show2Transform() : TimeVaryingTransform(false, false)
+    {
+        // Create our GUI proxy
+        gui = new GUIProxy();
+        // Move it to the main thread, this means signals we send to it will
+        // be run in the main thread, which is hopefully in an event loop
+        gui->moveToThread(QApplication::instance()->thread());
+        // Connect our signals to the proxy's slots
+        connect(this, SIGNAL(needWindow()), gui, SLOT(createWindow()), Qt::BlockingQueuedConnection);
+        connect(this, SIGNAL(updateImage(QPixmap)), gui,SLOT(showImage(QPixmap)));
+    }
+
+    ~Show2Transform()
+    {
+        delete gui;
+    }
+
+    void train(const TemplateList &data) { (void) data; }
+
+    void project(const TemplateList &src, TemplateList &dst) const
+    {
+        Transform * non_const = (Show2Transform *) this;
+        non_const->projectUpdate(src,dst);
+    }
+
+    void projectUpdate(const TemplateList &src, TemplateList &dst)
+    {
+        dst = src;
+
+        if (src.empty())
+            return;
+
+        foreach (const Template & t, src) {
+            foreach(const cv::Mat & m, t) {
+                qImageBuffer = toQImage(m);
+                displayBuffer.convertFromImage(qImageBuffer);
+                // Emit an explicit  copy of our pixmap so that the pixmap used
+                // by the main thread isn't damaged when we update displayBuffer
+                // later.
+                emit updateImage(displayBuffer.copy(displayBuffer.rect()));
+            }
+        }
+    }
+
+    void finalize(TemplateList & output)
+    {
+        (void) output;
+        // todo: hide window?
+    }
+
+    void init()
+    {
+        emit needWindow();
+    }
+
+protected:
+    GUIProxy * gui;
+    QImage qImageBuffer;
+    QPixmap displayBuffer;
+
+signals:
+    void needWindow();
+    void updateImage(const QPixmap & input);
+};
+
+BR_REGISTER(Transform, Show2Transform)
+
+} // namespace br
+
+#include "gui.moc"
diff --git a/sdk/plugins/hist.cpp b/sdk/plugins/hist.cpp
index 0372e54..996fae7 100644
--- a/sdk/plugins/hist.cpp
+++ b/sdk/plugins/hist.cpp
@@ -111,7 +111,7 @@ class RankTransform : public UntrainableTransform
         assert(m.channels() == 1);
         dst = Mat(m.rows, m.cols, CV_32FC1);
         typedef QPair<float,int> Tuple;
-        QList<Tuple> tuples = Common::Sort(OpenCVUtils::matrixToVector(m));
+        QList<Tuple> tuples = Common::Sort(OpenCVUtils::matrixToVector<float>(m));
 
         float prevValue = 0;
         int prevRank = 0;
diff --git a/sdk/plugins/integral.cpp b/sdk/plugins/integral.cpp
index d369cc9..3bfd356 100644
--- a/sdk/plugins/integral.cpp
+++ b/sdk/plugins/integral.cpp
@@ -28,7 +28,7 @@ BR_REGISTER(Transform, IntegralTransform)
 
 /*!
  * \ingroup transforms
- * \brief Sliding window feature extraction from a multi-channel intergral image.
+ * \brief Sliding window feature extraction from a multi-channel integral image.
  * \author Josh Klontz \cite jklontz
  */
 class IntegralSamplerTransform : public UntrainableTransform
@@ -38,9 +38,9 @@ class IntegralSamplerTransform : public UntrainableTransform
     Q_PROPERTY(float scaleFactor READ get_scaleFactor WRITE set_scaleFactor RESET reset_scaleFactor STORED false)
     Q_PROPERTY(float stepFactor READ get_stepFactor WRITE set_stepFactor RESET reset_stepFactor STORED false)
     Q_PROPERTY(int minSize READ get_minSize WRITE set_minSize RESET reset_minSize STORED false)
-    BR_PROPERTY(int, scales, 4)
+    BR_PROPERTY(int, scales, 5)
     BR_PROPERTY(float, scaleFactor, 1.5)
-    BR_PROPERTY(float, stepFactor, 0.25)
+    BR_PROPERTY(float, stepFactor, 0.75)
     BR_PROPERTY(int, minSize, 8)
 
     void project(const Template &src, Template &dst) const
@@ -53,22 +53,23 @@ class IntegralSamplerTransform : public UntrainableTransform
         const int rowStep = channels * m.cols;
 
         int descriptors = 0;
-        int currentSize = min(m.rows, m.cols)-1;
+        float idealSize = min(m.rows, m.cols)-1;
         for (int scale=0; scale<scales; scale++) {
-            descriptors += int(1+(m.rows-currentSize)/(currentSize*stepFactor)) *
-                           int(1+(m.cols-currentSize)/(currentSize*stepFactor));
-            currentSize /= scaleFactor;
-            if (currentSize < minSize)
-                break;
+            const int currentSize(idealSize);
+            descriptors += (1+(m.rows-currentSize-1)/int(idealSize*stepFactor)) *
+                           (1+(m.cols-currentSize-1)/int(idealSize*stepFactor));
+            idealSize /= scaleFactor;
+            if (idealSize < minSize) break;
         }
         Mat n(descriptors, channels, CV_32FC1);
 
         const qint32 *dataIn = (qint32*)m.data;
         float *dataOut = (float*)n.data;
-        currentSize = min(m.rows, m.cols)-1;
+        idealSize = min(m.rows, m.cols)-1;
         int index = 0;
         for (int scale=0; scale<scales; scale++) {
-            const int currentStep = currentSize * stepFactor;
+            const int currentSize(idealSize);
+            const int currentStep(idealSize*stepFactor);
             for (int i=currentSize; i<m.rows; i+=currentStep) {
                 for (int j=currentSize; j<m.cols; j+=currentStep) {
                     InputDescriptor a(dataIn+((i-currentSize)*rowStep+(j-currentSize)*channels), channels, 1);
@@ -80,10 +81,13 @@ class IntegralSamplerTransform : public UntrainableTransform
                     index++;
                 }
             }
-            currentSize /= scaleFactor;
-            if (currentSize < minSize)
-                break;
+            idealSize /= scaleFactor;
+            if (idealSize < minSize) break;
         }
+
+        if (descriptors != index)
+            qFatal("Allocated %d descriptors but computed %d.", descriptors, index);
+
         dst.m() = n;
     }
 };
@@ -133,38 +137,51 @@ class WordWiseTransform : public Transform
     Q_OBJECT
     Q_PROPERTY(br::Transform* getWords READ get_getWords WRITE set_getWords RESET reset_getWords)
     Q_PROPERTY(br::Transform* byWord READ get_byWord WRITE set_byWord RESET reset_byWord)
+    Q_PROPERTY(int numWords READ get_numWords WRITE set_numWords RESET reset_numWords)
     BR_PROPERTY(br::Transform*, getWords, NULL)
     BR_PROPERTY(br::Transform*, byWord, NULL)
+    BR_PROPERTY(int, numWords, 0)
 
     void train(const TemplateList &data)
     {
         getWords->train(data);
-        TemplateList words;
-        getWords->project(data, words);
+        TemplateList bins;
+        getWords->project(data, bins);
 
-        const int columns = data.first().m().cols;
-        int numWords = 0;
-        foreach (const Template &t, words) {
+        numWords = 0;
+        foreach (const Template &t, bins) {
             double minVal, maxVal;
             minMaxLoc(t, &minVal, &maxVal);
             numWords = max(numWords, int(maxVal)+1);
         }
 
-        QVector<int> wordCounts(numWords, 0);
-        foreach (const Template &t, words) {
-            const Mat &m = t.m();
-            for (int i=0; i<m.rows; i++)
-                wordCounts[m.at<uchar>(i,0)]++;
-        }
-
-        QVector<Mat> trainingWords(numWords);
-        for (int i=0; i<numWords; i++)
-            trainingWords[i] = Mat(wordCounts[i], columns, CV_8UC1);
+        TemplateList reworded; reworded.reserve(data.size());
+        foreach (const Template &t, data)
+            reworded.append(reword(t));
+        byWord->train(reworded);
     }
 
     void project(const Template &src, Template &dst) const
     {
-        (void) src; (void) dst;
+        byWord->project(reword(src), dst);
+    }
+
+    Template reword(const Template &src) const
+    {
+        Template words;
+        getWords->project(src, words);
+        QVector<int> wordCounts(numWords, 0);
+        for (int i=0; i<words.m().rows; i++)
+            wordCounts[words.m().at<uchar>(i,0)]++;
+        Template reworded(src.file); reworded.reserve(numWords);
+        for (int i=0; i<numWords; i++)
+            reworded.append(Mat(wordCounts[i], src.m().cols, src.m().type()));
+        QVector<int> indicies(numWords, 0);
+        for (int i=0; i<src.m().rows; i++) {
+            const int word = words.m().at<uchar>(i,0);
+            src.m().row(i).copyTo(reworded[word].row(indicies[word]++));
+        }
+        return reworded;
     }
 };
 
diff --git a/sdk/plugins/meta.cpp b/sdk/plugins/meta.cpp
index 1561d5d..5962aa0 100644
--- a/sdk/plugins/meta.cpp
+++ b/sdk/plugins/meta.cpp
@@ -14,6 +14,7 @@
  * limitations under the License.                                            *
  * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
+#include <QFutureSynchronizer>
 #include <QtConcurrentRun>
 #include <openbr_plugin.h>
 
@@ -60,76 +61,6 @@ static void _train(Transform *transform, const TemplateList *data)
     transform->train(*data);
 }
 
-// For handling progress feedback
-static int depth = 0;
-
-static void acquireStep()
-{
-    if (depth == 0) {
-        Globals->currentStep = 0;
-        Globals->totalSteps = 1;
-    }
-    depth++;
-}
-
-static void releaseStep()
-{
-    depth--;
-    Globals->currentStep = floor(Globals->currentStep * pow(10.0, double(depth))) / pow(10.0, double(depth));
-    if (depth == 0)
-        Globals->totalSteps = 0;
-}
-
-static void incrementStep()
-{
-    Globals->currentStep += 1.0 / pow(10.0, double(depth));
-}
-
-/*!
- * \brief A MetaTransform that aggregates some sub-transforms
- */
-class BR_EXPORT CompositeTransform : public TimeVaryingTransform
-{
-    Q_OBJECT
-
-public:
-    Q_PROPERTY(QList<br::Transform*> transforms READ get_transforms WRITE set_transforms RESET reset_transforms)
-    BR_PROPERTY(QList<br::Transform*>, transforms, QList<br::Transform*>())
-
-    virtual void project(const Template &src, Template &dst) const
-    {
-        if (timeVarying()) qFatal("No const project defined for time-varying transform");
-        _project(src, dst);
-    }
-
-    virtual void project(const TemplateList &src, TemplateList &dst) const
-    {
-        if (timeVarying()) qFatal("No const project defined for time-varying transform");
-        _project(src, dst);
-    }
-
-    bool timeVarying() const { return isTimeVarying; }
-
-    void init()
-    {
-        isTimeVarying = false;
-        foreach (const br::Transform *transform, transforms) {
-            if (transform->timeVarying()) {
-                isTimeVarying = true;
-                break;
-            }
-        }
-    }
-
-protected:
-    bool isTimeVarying;
-
-    virtual void _project(const Template & src, Template & dst) const = 0;
-    virtual void _project(const TemplateList & src, TemplateList & dst) const = 0;
-
-    CompositeTransform() : TimeVaryingTransform(false) {}
-};
-
 /*!
  * \ingroup Transforms
  * \brief Transforms in series.
@@ -146,16 +77,13 @@ class PipeTransform : public CompositeTransform
 
     void train(const TemplateList &data)
     {
-        acquireStep();
-
         TemplateList copy(data);
         for (int i=0; i<transforms.size(); i++) {
+            fprintf(stderr, "%s training... ", qPrintable(transforms[i]->objectName()));
             transforms[i]->train(copy);
+            fprintf(stderr, "projecting...\n");
             copy >> *transforms[i];
-            incrementStep();
         }
-
-        releaseStep();
     }
 
     void backProject(const Template &dst, Template &src) const
@@ -284,6 +212,38 @@ BR_REGISTER(Transform, ExpandTransform)
 
 /*!
  * \ingroup transforms
+ * \brief It's like the opposite of ExpandTransform, but not really
+ * \author Charles Otto \cite caotto
+ *
+ * Given a set of templatelists as input, concatenate them onto a single Template
+ */
+class ContractTransform : public UntrainableMetaTransform
+{
+    Q_OBJECT
+
+    virtual void project(const TemplateList &src, TemplateList &dst) const
+    {
+        //dst = Expanded(src);
+        Template out;
+
+        foreach (const Template & t, src) {
+            out.merge(t);
+        }
+        dst.clear();
+        dst.append(out);
+    }
+
+    virtual void project(const Template & src, Template & dst) const
+    {
+        qFatal("this has gone bad");
+        (void) src; (void) dst;
+    }
+};
+
+BR_REGISTER(Transform, ContractTransform)
+
+/*!
+ * \ingroup transforms
  * \brief Transforms in parallel.
  * \author Josh Klontz \cite jklontz
  *
@@ -297,13 +257,12 @@ class ForkTransform : public CompositeTransform
 
     void train(const TemplateList &data)
     {
-        QList< QFuture<void> > futures;
-        const bool threaded = Globals->parallelism && (transforms.size() > 1);
+        QFutureSynchronizer<void> futures;
         for (int i=0; i<transforms.size(); i++) {
-            if (threaded) futures.append(QtConcurrent::run(_train, transforms[i], &data));
-            else                                           _train (transforms[i], &data);
+            if (Globals->parallelism) futures.addFuture(QtConcurrent::run(_train, transforms[i], &data));
+            else                                                          _train (transforms[i], &data);
         }
-        if (threaded) Globals->trackFutures(futures);
+        futures.waitForFinished();
     }
 
     void backProject(const Template &dst, Template &src) const {Transform::backProject(dst, src);}
@@ -626,6 +585,15 @@ public:
     // Process the single elemnt templates in parallel if parallelism is enabled.
     void project(const TemplateList &src, TemplateList &dst) const
     {
+        // Little ugly, but if we own a timeVaryingTransform and this gets called
+        // cast off the const modifier and use projectUpdate. This allows us to
+        // act as a single point of entry.
+        if (transform->timeVarying())
+        {
+            DistributeTemplateTransform * non_const = (DistributeTemplateTransform *) this;
+            non_const->projectUpdate(src,dst);
+            return;
+        }
         // Pre-allocate output for each template
         QList<TemplateList> output_buffer;
         output_buffer.reserve(src.size());
@@ -639,22 +607,26 @@ public:
             output_buffer.append(TemplateList());
         }
 
-        QList< QFuture<void> > futures;
-        futures.reserve(src.size());
+        QFutureSynchronizer<void> futures;
         for (int i=0; i<src.size(); i++) {
             input_buffer[i].append(src[i]);
-            if (Globals->parallelism)
-                futures.append(QtConcurrent::run(_projectList, transform, &input_buffer[i], &output_buffer[i]));
-            else
-                _projectList(transform, &input_buffer[i], &output_buffer[i]);
+            if (Globals->parallelism) futures.addFuture(QtConcurrent::run(_projectList, transform, &input_buffer[i], &output_buffer[i]));
+            else                                                          _projectList( transform, &input_buffer[i], &output_buffer[i]);
         }
-
-        if (Globals->parallelism)
-            Globals->trackFutures(futures);
+        futures.waitForFinished();
 
         for (int i=0; i<src.size(); i++) dst.append(output_buffer[i]);
     }
 
+    void projectUpdate(const TemplateList &src, TemplateList &dst)
+    {
+        if (!transform->timeVarying()) {
+            this->project(src, dst);
+            return;
+        }
+        this->transform->projectUpdate(src, dst);
+    }
+
 
     private:
 
diff --git a/sdk/plugins/normalize.cpp b/sdk/plugins/normalize.cpp
index 4300bf4..a924e07 100644
--- a/sdk/plugins/normalize.cpp
+++ b/sdk/plugins/normalize.cpp
@@ -14,6 +14,7 @@
  * limitations under the License.                                            *
  * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
+#include <QFutureSynchronizer>
 #include <QtConcurrentRun>
 #include <opencv2/imgproc/imgproc.hpp>
 #include <opencv2/highgui/highgui.hpp>
@@ -22,6 +23,7 @@
 
 #include "core/common.h"
 #include "core/opencvutils.h"
+#include "core/qtutils.h"
 
 using namespace cv;
 
@@ -119,18 +121,16 @@ private:
             bv.push_back(Mat(1, dims, CV_64FC1));
         }
 
-        QList< QFuture<void> > futures; futures.reserve(dims);
+        QFutureSynchronizer<void> futures;
         const bool parallel = (data.size() > 1000) && Globals->parallelism;
-
         for (size_t c = 0; c < mv.size(); c++) {
             for (int i=0; i<dims; i++)
-                if (parallel) futures.append(QtConcurrent::run(_train, method, mv[c], &av[c], &bv[c], i));
-                else                                           _train (method, mv[c], &av[c], &bv[c], i);
+                if (parallel) futures.addFuture(QtConcurrent::run(_train, method, mv[c], &av[c], &bv[c], i));
+                else                                              _train (method, mv[c], &av[c], &bv[c], i);
             av[c] = av[c].reshape(1, data.first().m().rows);
             bv[c] = bv[c].reshape(1, data.first().m().rows);
         }
-
-        if (parallel) Globals->trackFutures(futures);
+        futures.waitForFinished();
 
         merge(av, a);
         merge(bv, b);
@@ -224,6 +224,30 @@ class RowWiseMeanCenterTransform : public Transform
 
 BR_REGISTER(Transform, RowWiseMeanCenterTransform)
 
+/*!
+ * \ingroup transforms
+ * \brief dst=sqrt(norm_L1(src)) proposed as RootSIFT in \cite Arandjelovic12
+ * \author Josh Klontz \cite jklontz
+ */
+class RootNormTransform : public UntrainableTransform
+{
+    Q_OBJECT
+
+    void project(const Template &src, Template &dst) const
+    {
+        const Mat &m = src;
+        dst.m() = Mat(m.rows, m.cols, m.type());
+        for (int i=0; i<m.rows; i++) {
+            Mat temp;
+            cv::normalize(m.row(i), temp, 1, 0, NORM_L1);
+            cv::sqrt(temp, temp);
+            temp.copyTo(dst.m().row(i));
+        }
+    }
+};
+
+BR_REGISTER(Transform, RootNormTransform)
+
 } // namespace br
 
 #include "normalize.moc"
diff --git a/sdk/plugins/output.cpp b/sdk/plugins/output.cpp
index 44c2a0d..ed48ca8 100644
--- a/sdk/plugins/output.cpp
+++ b/sdk/plugins/output.cpp
@@ -176,7 +176,7 @@ class rrOutput : public MatrixOutput
             if (!byLine) files.append(queryFiles[i]);
 
             typedef QPair<float,int> Pair;
-            foreach (const Pair &pair, Common::Sort(OpenCVUtils::matrixToVector(data.row(i)), true, limit)) {
+            foreach (const Pair &pair, Common::Sort(OpenCVUtils::matrixToVector<float>(data.row(i)), true, limit)) {
                 if (pair.first < threshold) break;
                 File target = targetFiles[pair.second];
                 target.set("Score", QString::number(pair.first));
@@ -278,7 +278,7 @@ class rankOutput : public MatrixOutput
         for (int i=0; i<queryFiles.size(); i++) {
             typedef QPair<float,int> Pair;
             int rank = 1;
-            foreach (const Pair &pair, Common::Sort(OpenCVUtils::matrixToVector(data.row(i)), true)) {
+            foreach (const Pair &pair, Common::Sort(OpenCVUtils::matrixToVector<float>(data.row(i)), true)) {
                 if(targetFiles[pair.second].get<QString>("Label") == queryFiles[i].get<QString>("Label")) {
                     ranks.append(rank);
                     positions.append(pair.second);
diff --git a/sdk/plugins/quantize.cpp b/sdk/plugins/quantize.cpp
index 416289b..6624a17 100644
--- a/sdk/plugins/quantize.cpp
+++ b/sdk/plugins/quantize.cpp
@@ -14,6 +14,8 @@
  * limitations under the License.                                            *
  * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
+#include <QFutureSynchronizer>
+#include <QtConcurrentRun>
 #include <openbr_plugin.h>
 
 #include "core/opencvutils.h"
@@ -109,6 +111,145 @@ class PackTransform : public UntrainableTransform
 
 BR_REGISTER(Transform, PackTransform)
 
+QVector<Mat> ProductQuantizationLUTs;
+
+/*!
+ * \ingroup distances
+ * \brief Distance in a product quantized space \cite jegou11
+ * \author Josh Klontz
+ */
+class ProductQuantizationDistance : public Distance
+{
+    Q_OBJECT
+    Q_PROPERTY(bool bayesian READ get_bayesian WRITE set_bayesian RESET reset_bayesian STORED false)
+    BR_PROPERTY(bool, bayesian, false)
+
+    float compare(const Template &a, const Template &b) const
+    {
+        float distance = 0;
+        for (int i=0; i<a.size(); i++) {
+            const int elements = a[i].total();
+            const uchar *aData = a[i].data;
+            const uchar *bData = b[i].data;
+            const float *lut = (const float*)ProductQuantizationLUTs[i].data;
+            for (int j=0; j<elements; j++)
+                 distance += lut[i*256*256 + aData[j]*256+bData[j]];
+        }
+        if (!bayesian) distance = -log(distance+1);
+        return distance;
+    }
+};
+
+BR_REGISTER(Distance, ProductQuantizationDistance)
+
+/*!
+ * \ingroup transforms
+ * \brief Product quantization \cite jegou11
+ * \author Josh Klontz \cite jklontz
+ */
+class ProductQuantizationTransform : public Transform
+{
+    Q_OBJECT
+    Q_PROPERTY(int n READ get_n WRITE set_n RESET reset_n STORED false)
+    Q_PROPERTY(bool bayesian READ get_bayesian WRITE set_bayesian RESET reset_bayesian STORED false)
+    BR_PROPERTY(int, n, 2)
+    BR_PROPERTY(bool, bayesian, false)
+
+    int index;
+    QList<Mat> centers;
+
+public:
+    ProductQuantizationTransform()
+    {
+        index = ProductQuantizationLUTs.size();
+        ProductQuantizationLUTs.append(Mat());
+    }
+
+private:
+    static double likelihoodRatio(const QPair<int,int> &totals, const QList<int> &targets, const QList<int> &queries)
+    {
+        int positives = 0, negatives = 0;
+        foreach (int t, targets)
+            foreach (int q, queries)
+                if (t == q) positives++;
+                else        negatives++;
+        return log(float(positives)/float(totals.first)) / (float(negatives)/float(totals.second));
+    }
+
+    void _train(const Mat &data, const QPair<int,int> &totals, Mat &lut, int i, const QList<int> &templateLabels)
+    {
+        Mat labels, center;
+        kmeans(data.colRange(i*n,(i+1)*n), 256, labels, TermCriteria(TermCriteria::MAX_ITER, 10, 0), 3, KMEANS_PP_CENTERS, center);
+        QList<int> clusterLabels = OpenCVUtils::matrixToVector<int>(labels);
+        QHash< int, QList<int> > clusters; // QHash<clusterLabel, QList<templateLabel> >
+        for (int i=0; i<clusterLabels.size(); i++)
+            clusters[clusterLabels[i]].append(templateLabels[i]);
+
+        for (int j=0; j<256; j++)
+            for (int k=0; k<256; k++)
+                lut.at<float>(i,j*256+k) = bayesian ? likelihoodRatio(totals, clusters[j], clusters[k]) :
+                                                      norm(center.row(j), center.row(k), NORM_L2);
+        centers[i] = center;
+    }
+
+    void train(const TemplateList &src)
+    {
+        Mat data = OpenCVUtils::toMat(src.data());
+        if (data.cols % n != 0) qFatal("Expected dimensionality to be divisible by n.");
+        const QList<int> templateLabels = src.labels<int>();
+        int totalPositives = 0, totalNegatives = 0;
+        for (int i=0; i<templateLabels.size(); i++)
+            for (int j=0; j<templateLabels.size(); j++)
+                if (templateLabels[i] == templateLabels[j]) totalPositives++;
+                else                                        totalNegatives++;
+        QPair<int,int> totals(totalPositives, totalNegatives);
+
+        Mat &lut = ProductQuantizationLUTs[index];
+        lut = Mat(data.cols/n, 256*256, CV_32FC1);
+
+        for (int i=0; i<lut.rows; i++)
+            centers.append(Mat());
+
+        QFutureSynchronizer<void> futures;
+        for (int i=0; i<lut.rows; i++) {
+            if (Globals->parallelism) futures.addFuture(QtConcurrent::run(this, &ProductQuantizationTransform::_train, data, totals, lut, i, templateLabels));
+            else                                                                                               _train (data, totals, lut, i, templateLabels);
+        }
+        futures.waitForFinished();
+    }
+
+    void project(const Template &src, Template &dst) const
+    {
+        Mat m = src.m().reshape(1, 1);
+        dst = Mat(1, m.cols/n, CV_8UC1);
+        for (int i=0; i<dst.m().cols; i++) {
+            int bestIndex = 0;
+            double bestDistance = std::numeric_limits<double>::max();
+            Mat m_i = m.colRange(i*n, (i+1)*n);
+            for (int j=0; j<256; j++) {
+                double distance = norm(m_i, centers[index].row(j), NORM_L2);
+                if (distance < bestDistance) {
+                    bestDistance = distance;
+                    bestIndex = j;
+                }
+            }
+            dst.m().at<uchar>(0,i) = bestIndex;
+        }
+    }
+
+    void store(QDataStream &stream) const
+    {
+        stream << centers << ProductQuantizationLUTs[index];
+    }
+
+    void load(QDataStream &stream)
+    {
+        stream >> centers >> ProductQuantizationLUTs[index];
+    }
+};
+
+BR_REGISTER(Transform, ProductQuantizationTransform)
+
 } // namespace br
 
 #include "quantize.moc"
diff --git a/sdk/plugins/regions.cpp b/sdk/plugins/regions.cpp
index 6248879..541f963 100644
--- a/sdk/plugins/regions.cpp
+++ b/sdk/plugins/regions.cpp
@@ -120,6 +120,7 @@ class MergeTransform : public UntrainableMetaTransform
 
     void project(const Template &src, Template &dst) const
     {
+        dst.file = src.file;
         std::vector<Mat> mv;
         foreach (const Mat &m, src)
             mv.push_back(m);
diff --git a/sdk/plugins/sentence.cpp b/sdk/plugins/sentence.cpp
new file mode 100644
index 0000000..4a60897
--- /dev/null
+++ b/sdk/plugins/sentence.cpp
@@ -0,0 +1,93 @@
+#include <stdint.h>
+#include <openbr_plugin.h>
+
+using namespace cv;
+
+namespace br
+{
+
+/*!
+ * \ingroup transforms
+ * \brief Ordered words
+ * \author Josh Klontz \cite jklontz
+ */
+class SentenceTransform : public UntrainableMetaTransform
+{
+    Q_OBJECT
+
+    void project(const Template &src, Template &dst) const
+    {
+        QByteArray sentence;
+        QDataStream stream(&sentence, QIODevice::WriteOnly);
+        for (int i=0; i<src.size(); i++) {
+            const Mat &m = src[i];
+            if (!m.data) continue;
+            stream.writeRawData((const char*)&i, 4);
+            stream.writeRawData((const char*)&m.rows, 4);
+            stream.writeRawData((const char*)&m.cols, 4);
+            stream.writeRawData((const char*)m.data, 4*m.rows*m.cols);
+        }
+        dst.file = src.file;
+        dst.m() = Mat(1, sentence.size(), CV_8UC1, sentence.data()).clone();
+    }
+};
+
+BR_REGISTER(Transform, SentenceTransform)
+
+/*!
+ * \ingroup distances
+ * \brief Distance between sentences
+ * \author Josh Klontz \cite jklontz
+ */
+class SentenceSimilarityDistance : public Distance
+{
+    Q_OBJECT
+
+    float compare(const Template &a, const Template &b) const
+    {
+        uchar *aBuffer = a.m().data;
+        uchar *bBuffer = b.m().data;
+        const uchar *aEnd = aBuffer + a.m().cols;
+        const uchar *bEnd = bBuffer + b.m().cols;
+
+        int32_t aWord, bWord, aRows, bRows, aColumns, bColumns;
+        float *aData, *bData;
+        aWord = aRows = aColumns = -2;
+        bWord = bRows = bColumns = -1;
+        aData = bData = NULL;
+
+        float distance = 0;
+        int comparisons = 0;
+        while (true) {
+            if (aWord < bWord) {
+                if (aBuffer == aEnd) return distance == 0 ? -std::numeric_limits<float>::max() : comparisons / distance;
+                aWord = *reinterpret_cast<int32_t*>(aBuffer);
+                aRows = *reinterpret_cast<int32_t*>(aBuffer+4);
+                aColumns = *reinterpret_cast<int32_t*>(aBuffer+8);
+                aData = reinterpret_cast<float*>(aBuffer+12);
+                aBuffer += 12 + 4*aRows*aColumns;
+            } else if (bWord < aWord) {
+                if (bBuffer == bEnd) return comparisons == 0 ? -std::numeric_limits<float>::max() : comparisons / distance;
+                bWord = *reinterpret_cast<int32_t*>(bBuffer);
+                bRows = *reinterpret_cast<int32_t*>(bBuffer+4);
+                bColumns = *reinterpret_cast<int32_t*>(bBuffer+8);
+                bData = reinterpret_cast<float*>(bBuffer+12);
+                bBuffer += 12 + 4*bRows*bColumns;
+            } else {
+                for (int i=0; i<aRows; i++)
+                    for (int j=0; j<bRows; j++)
+                        for (int k=0; k<aColumns; k++)
+                            distance += pow(aBuffer[i*aColumns+k] - bBuffer[j*bColumns+k], 2.f);
+                comparisons += aRows * bRows * aColumns;
+                aWord = -2;
+                bWord = -1;
+            }
+        }
+    }
+};
+
+BR_REGISTER(Distance, SentenceSimilarityDistance)
+
+} // namespace br
+
+#include "sentence.moc"
diff --git a/sdk/plugins/stream.cpp b/sdk/plugins/stream.cpp
new file mode 100644
index 0000000..b0b0eeb
--- /dev/null
+++ b/sdk/plugins/stream.cpp
@@ -0,0 +1,790 @@
+#include <openbr_plugin.h>
+#include <QReadWriteLock>
+#include <QWaitCondition>
+#include <QThreadPool>
+#include <QSemaphore>
+#include <QMap>
+
+#include "core/common.h"
+#include "core/opencvutils.h"
+#include "core/qtutils.h"
+
+#include "opencv/highgui.h"
+
+#include <iostream>
+
+using namespace cv;
+
+namespace br
+{
+
+class FrameData
+{
+public:
+    int sequenceNumber;
+    TemplateList data;
+};
+
+// A buffer shared between adjacent processing stages in a stream
+class SharedBuffer
+{
+public:
+    SharedBuffer() {}
+    virtual ~SharedBuffer() {}
+
+    virtual void addItem(FrameData * input)=0;
+
+    virtual FrameData * getItem()=0;
+
+    virtual void stoppedInput() =0;
+    virtual void startInput() = 0;
+};
+
+// For 1 - n boundaries, a buffer class with a single shared buffer, a mutex
+// is used to serialize all access to the buffer.
+class SingleBuffer : public SharedBuffer
+{
+public:
+    SingleBuffer() { no_input = false; }
+
+    void stoppedInput()
+    {
+        QMutexLocker bufferLock(&bufferGuard);
+        no_input = true;
+        // Release anything waiting for input items.
+        availableInput.wakeAll();
+    }
+
+    // There will be more input
+    void startInput()
+    {
+        QMutexLocker bufferLock(&bufferGuard);
+        no_input = false;
+    }
+
+    void addItem(FrameData * input)
+    {
+        QMutexLocker bufferLock(&bufferGuard);
+
+        buffer.append(input);
+
+        availableInput.wakeOne();
+    }
+
+    FrameData * getItem()
+    {
+        QMutexLocker bufferLock(&bufferGuard);
+
+        if (buffer.empty()) {
+            // If no further items will come we are done here
+            if (no_input)
+                return NULL;
+            // Wait for an item
+            availableInput.wait(&bufferGuard);
+        }
+
+        // availableInput was signalled, but the buffer is still empty? We're done here.
+        if (buffer.empty())
+            return NULL;
+
+        FrameData * output = buffer.first();
+        buffer.removeFirst();
+        return output;
+    }
+
+private:
+    QMutex bufferGuard;
+    QWaitCondition availableInput;
+    bool no_input;
+
+    QList<FrameData *> buffer;
+};
+
+// for n - 1 boundaries, multiple threads call addItem, the frames are
+// sequenced based on FrameData::sequence_number, and calls to getItem
+// receive them in that order
+class SequencingBuffer : public SharedBuffer
+{
+public:
+    SequencingBuffer()
+    {
+        no_input = false;
+        next_target = 0;
+    }
+
+    void stoppedInput()
+    {
+        QMutexLocker bufferLock(&bufferGuard);
+        no_input = true;
+        // Release anything waiting for input items.
+        availableInput.wakeAll();
+    }
+
+    // There will be more input
+    void startInput()
+    {
+        QMutexLocker bufferLock(&bufferGuard);
+        no_input = false;
+    }
+
+    void addItem(FrameData * input)
+    {
+        QMutexLocker bufferLock(&bufferGuard);
+
+        buffer.insert(input->sequenceNumber, input);
+
+        if (input->sequenceNumber == next_target) {
+            availableInput.wakeOne();
+        }
+    }
+
+    FrameData * getItem()
+    {
+        QMutexLocker bufferLock(&bufferGuard);
+
+        if (buffer.empty() || buffer.begin().key() != this->next_target) {
+            if (buffer.empty() && no_input) {
+                next_target = 0;
+                return NULL;
+            }
+            availableInput.wait(&bufferGuard);
+        }
+
+        // availableInput was signalled, but the buffer is empty? We're done here.
+        if (buffer.empty()) {
+            next_target = 0;
+            return NULL;
+        }
+
+        QMap<int, FrameData *>::Iterator result = buffer.begin();
+        //next_target++;
+        if (next_target != result.value()->sequenceNumber) {
+            qWarning("mismatched targets!");
+        }
+
+        next_target = next_target + 1;
+
+        FrameData * output = result.value();
+        buffer.erase(result);
+        return output;
+    }
+
+private:
+    QMutex bufferGuard;
+    QWaitCondition availableInput;
+    bool no_input;
+
+    int next_target;
+
+    QMap<int, FrameData *> buffer;
+};
+
+// For 1 - 1 boundaries, a double buffering scheme
+// Producer/consumer read/write from separate buffers, and switch if their
+// buffer runs out/overflows. Synchronization is handled by a read/write lock
+// threads are "reading" if they are adding to/removing from their individual
+// buffer, and writing if they access or swap with the other buffer.
+class DoubleBuffer : public SharedBuffer
+{
+public:
+    DoubleBuffer()
+    {
+        inputBuffer = &buffer1;
+        outputBuffer = &buffer2;
+    }
+
+    void stoppedInput()
+    {
+        QWriteLocker bufferLock(&bufferGuard);
+        no_input = true;
+        // Release anything waiting for input items.
+        availableInput.wakeAll();
+    }
+
+    // There will be more input
+    void startInput()
+    {
+        QWriteLocker bufferLock(&bufferGuard);
+        no_input = false;
+    }
+
+    // called from the producer thread
+    void addItem(FrameData * input)
+    {
+        QReadLocker readLock(&bufferGuard);
+        inputBuffer->append(input);
+        availableInput.wakeOne();
+    }
+
+    // Called from the consumer thread
+    FrameData * getItem() {
+        QReadLocker readLock(&bufferGuard);
+
+        // There is something for us to get
+        if (!outputBuffer->empty()) {
+            FrameData * output = outputBuffer->first();
+            outputBuffer->removeFirst();
+            return output;
+        }
+
+        // Outputbuffer is empty, try to swap with the input buffer, we need a
+        // write lock to do that.
+        readLock.unlock();
+        QWriteLocker writeLock(&bufferGuard);
+
+        // Nothing on the input buffer either?
+        if (inputBuffer->empty()) {
+            // If nothing else is coming, return null
+            if (no_input)
+                return NULL;
+            //otherwise, wait on the input buffer
+            availableInput.wait(&bufferGuard);
+            // Did we get woken up because no more input is coming? if so
+            // we're done here
+            if (no_input && inputBuffer->empty())
+                return NULL;
+        }
+
+        // input buffer is non-empty, so swap the buffers
+        std::swap(inputBuffer, outputBuffer);
+
+        // Return a frame
+        FrameData * output = outputBuffer->first();
+        outputBuffer->removeFirst();
+        return output;
+    }
+
+private:
+    // The read-write lock. The thread adding to this buffer can add
+    // to the current input buffer if it has a read lock. The thread
+    // removing from this buffer can remove things from the current
+    // output buffer if it has a read lock, or swap the buffers if it
+    // has a write lock.
+    // Checking/modifying no_input requires a write lock.
+    QReadWriteLock bufferGuard;
+    QWaitCondition availableInput;
+    bool no_input;
+
+    // The buffer that is currently being added to
+    QList<FrameData *> * inputBuffer;
+    // The buffer that is currently being removed from
+    QList<FrameData *> * outputBuffer;
+
+    // The buffers pointed at by inputBuffer/outputBuffer
+    QList<FrameData *> buffer1;
+    QList<FrameData *> buffer2;
+};
+
+
+// Interface for sequentially getting data from some data source.
+// Initialized off of a template, can represent a video file (stored in the template's filename)
+// or a set of images already loaded into memory stored as multiple matrices in an input template.
+class DataSource
+{
+public:
+    DataSource(int maxFrames=100)
+    {
+        for (int i=0; i < maxFrames;i++)
+        {
+            allFrames.addItem(new FrameData());
+        }
+        allFrames.startInput();
+    }
+
+    virtual ~DataSource()
+    {
+        allFrames.stoppedInput();
+        while (true)
+        {
+            FrameData * frame = allFrames.getItem();
+            if (frame == NULL)
+                break;
+            delete frame;
+        }
+    }
+
+    FrameData * getFrame()
+    {
+        FrameData * aFrame = allFrames.getItem();
+        aFrame->data.clear();
+        aFrame->sequenceNumber = -1;
+
+        bool res = getNext(*aFrame);
+        if (!res) {
+            allFrames.addItem(aFrame);
+            return NULL;
+        }
+        return aFrame;
+    }
+
+    void returnFrame(FrameData * inputFrame)
+    {
+        allFrames.addItem(inputFrame);
+    }
+
+    virtual void close() = 0;
+    virtual bool open(Template & output) = 0;
+    virtual bool isOpen() = 0;
+
+    virtual bool getNext(FrameData & input) = 0;
+
+protected:
+    DoubleBuffer allFrames;
+};
+
+// Read a video frame by frame using cv::VideoCapture
+class VideoDataSource : public DataSource
+{
+public:
+    VideoDataSource(int maxFrames) : DataSource(maxFrames) {}
+
+    bool open(Template &input)
+    {
+        next_idx = 0;
+        basis = input;
+        video.open(input.file.name.toStdString());
+        return video.isOpened();
+    }
+
+    bool isOpen() { return video.isOpened(); }
+
+    void close() { video.release(); }
+
+private:
+    bool getNext(FrameData & output)
+    {
+        if (!isOpen())
+            return false;
+
+        output.data.append(Template(basis.file));
+        output.data.last().append(cv::Mat());
+
+        output.sequenceNumber = next_idx;
+        next_idx++;
+
+        bool res = video.read(output.data.last().last());
+        if (!res) {
+            return false;
+        }
+        return true;
+    }
+
+    cv::VideoCapture video;
+    Template basis;
+    int next_idx;
+};
+
+// Given a template as input, return its matrices one by one on subsequent calls
+// to getNext
+class TemplateDataSource : public DataSource
+{
+public:
+    TemplateDataSource(int maxFrames) : DataSource(maxFrames)
+    {
+        current_idx = INT_MAX;
+    }
+
+    bool open(Template &input)
+    {
+        basis = input;
+        current_idx = 0;
+        next_sequence = 0;
+        return isOpen();
+    }
+
+    bool isOpen() { return current_idx < basis.size() ; }
+
+    void close()
+    {
+        current_idx = INT_MAX;
+        basis.clear();
+    }
+
+private:
+    bool getNext(FrameData & output)
+    {
+        if (!isOpen())
+            return false;
+
+        output.data.append(basis[current_idx]);
+        current_idx++;
+
+        output.sequenceNumber = next_sequence;
+        next_sequence++;
+
+        return true;
+    }
+
+    Template basis;
+    int current_idx;
+    int next_sequence;
+};
+
+// Given a template as input, create a VideoDataSource or a TemplateDataSource
+// depending on whether or not it looks like the input template has already
+// loaded frames into memory.
+class DataSourceManager : public DataSource
+{
+public:
+    DataSourceManager()
+    {
+        actualSource = NULL;
+    }
+
+    ~DataSourceManager()
+    {
+        close();
+    }
+
+    void close()
+    {
+        if (actualSource) {
+            actualSource->close();
+            delete actualSource;
+            actualSource = NULL;
+        }
+    }
+
+    bool open(Template & input)
+    {
+        close();
+        bool open_res = false;
+        // Input has no matrices? Its probably a video that hasn't been loaded yet
+        if (input.empty()) {
+            actualSource = new VideoDataSource(0);
+            open_res = actualSource->open(input);
+        }
+        else {
+            // create frame dealer
+            actualSource = new TemplateDataSource(0);
+            open_res = actualSource->open(input);
+        }
+        if (!isOpen()) {
+            delete actualSource;
+            actualSource = NULL;
+            return false;
+        }
+        return true;
+    }
+
+    bool isOpen() { return !actualSource ? false : actualSource->isOpen(); }
+
+protected:
+    DataSource * actualSource;
+    bool getNext(FrameData & output)
+    {
+        return actualSource->getNext(output);
+    }
+
+};
+
+class ProcessingStage : public QRunnable
+{
+    friend class StreamTransform;
+public:
+    ProcessingStage(int nThreads = 1)
+    {
+        thread_count = nThreads;
+        activeThreads.release(thread_count);
+        setAutoDelete(false);
+    }
+
+    void markStart()
+    {
+        activeThreads.acquire();
+    }
+
+    void waitStop()
+    {
+        // Wait until all threads have stopped
+        activeThreads.acquire(thread_count);
+        activeThreads.release(thread_count);
+    }
+
+protected:
+    void markStop()
+    {
+        activeThreads.release();
+    }
+    QSemaphore activeThreads;
+    int thread_count;
+
+    SharedBuffer * inputBuffer;
+    SharedBuffer * outputBuffer;
+    Transform * transform;
+    int stage_id;
+
+public:
+    // We should start, and enter a wait on input data
+    void run()
+    {
+        markStart();
+        forever
+        {
+            FrameData * currentItem = inputBuffer->getItem();
+            if (currentItem == NULL)
+                break;
+            // Project the input we got
+            transform->projectUpdate(currentItem->data);
+            // Add the result to the ouptut buffer
+            outputBuffer->addItem(currentItem);
+        }
+        markStop();
+    }
+};
+
+
+// No input buffer, instead we draw templates from some data source
+// Will be operated by the main thread for the stream
+class FirstStage : public ProcessingStage
+{
+public:
+    DataSourceManager dataSource;
+    // Start drawing frames from the datasource.
+    void run()
+    {
+        forever
+        {
+            //FrameData * aFrame = dataSource.getNext();
+            FrameData * aFrame = dataSource.getFrame();
+            if (aFrame == NULL)
+                break;
+            outputBuffer->addItem(aFrame);
+        }
+        this->markStop();
+    }
+};
+
+class LastStage : public ProcessingStage
+{
+public:
+    TemplateList getOutput()
+    {
+        return collectedOutput;
+    }
+
+private:
+    TemplateList collectedOutput;
+public:
+    DataSource * data;
+    void run()
+    {
+        forever
+        {
+            // Wait for input
+            FrameData * frame = inputBuffer->getItem();
+            if (frame == NULL)
+                break;
+            // Just put the item on collectedOutput
+            collectedOutput.append(frame->data);
+            // Return the frame to the input frame buffer
+            data->returnFrame(frame);
+        }
+        this->markStop();
+    }
+};
+
+class StreamTransform : public CompositeTransform
+{
+    Q_OBJECT
+    int threads_per_multi_stage;
+public:
+    void train(const TemplateList & data)
+    {
+        foreach(Transform * transform, transforms) {
+            transform->train(data);
+        }
+    }
+
+    bool timeVarying() const { return true; }
+
+    void project(const Template &src, Template &dst) const
+    {
+        (void) src; (void) dst;
+        qFatal("nope");
+    }
+    void project(const TemplateList & src, TemplateList & dst) const
+    {
+        (void) src; (void) dst;
+        qFatal("nope");
+    }
+
+    void projectUpdate(const Template &src, Template &dst)
+    {
+        (void) src; (void) dst;
+        qFatal("whatever");
+    }
+
+    // start processing
+    void projectUpdate(const TemplateList & src, TemplateList & dst)
+    {
+        if (src.size() != 1)
+            qFatal("Expected single template input to stream");
+
+        dst = src;
+        bool res = readStage.dataSource.open(dst[0]);
+        if (!res) {
+            qWarning("failed to stream template %s", qPrintable(dst[0].file.name));
+            return;
+        }
+
+        // Tell all buffers to expect input
+        for (int i=0; i < sharedBuffers.size(); i++) {
+            sharedBuffers[i]->startInput();
+        }
+
+        // Start our processing stages
+        for (int i=0; i < this->processingStages.size(); i++) {
+            int count = stage_variance[i] ? 1 : threads_per_multi_stage;
+            for (int j =0; j < count; j ++) processingThreads.start(processingStages[i]);
+        }
+
+        // Start the final stage
+        processingThreads.start(&collectionStage);
+
+        // Run the read stage ourselves
+        readStage.run();
+
+        // The read stage has stopped (since we ran the read stage).
+        // Step over the buffers, and call stoppedInput to tell the stage
+        // reading from each buffer that no more frames will be added after
+        // the current ones run out, then wait for the thread to finish.
+        for (int i =0; i < (sharedBuffers.size() - 1); i++) {
+            // Indicate that no more input will be available
+            sharedBuffers[i]->stoppedInput();
+
+            // Wait for the thread to finish.
+            this->processingStages[i]->waitStop();
+        }
+        // Wait for the collection stage to finish
+        sharedBuffers.last()->stoppedInput();
+        collectionStage.waitStop();
+
+        // dst is set to all output received by the final stage
+        dst = collectionStage.getOutput();
+    }
+
+    virtual void finalize(TemplateList & output)
+    {
+        (void) output;
+        // Not handling this yet -cao
+    }
+
+    // Create and link stages
+    void init()
+    {
+        int thread_count = 0;
+        threads_per_multi_stage = 4;
+        stage_variance.reserve(transforms.size());
+        foreach (const br::Transform *transform, transforms) {
+            stage_variance.append(transform->timeVarying());
+            thread_count += transform->timeVarying() ? 1 : threads_per_multi_stage;
+        }
+        if (transforms.isEmpty()) return;
+
+        // Set up the thread pool, 1 stage for each transform, as well as first
+        // and last stages, but the first stage is operated by the thread that
+        // calls project so the pool only needs nTransforms+1 total.
+        processingThreads.setMaxThreadCount(thread_count + 1);
+
+
+        // buffer 0 -- output buffer for the read stage, input buffer for
+        // first transform. Is that transform time-varying?
+        if (stage_variance[0])
+            sharedBuffers.append(new DoubleBuffer());
+        // If not, we can run multiple threads
+        else
+            sharedBuffers.append(new SingleBuffer());
+
+        readStage.outputBuffer = sharedBuffers.last();
+        readStage.stage_id = 0;
+
+        int next_stage_id = 1;
+
+        int lastBufferIdx = 0;
+        for (int i =0; i < transforms.size(); i++)
+        {
+            // Set up this stage
+            processingStages.append(new ProcessingStage(stage_variance[i] ? 1 : threads_per_multi_stage));
+
+            processingStages.last()->stage_id = next_stage_id++;
+            processingStages.last()->inputBuffer = sharedBuffers[lastBufferIdx];
+            lastBufferIdx++;
+
+            // This stage's output buffer, next stage's input buffer. If this is
+            // the last transform, the next stage is the (time varying) collection
+            // stage
+            bool next_variance = (i+1) < transforms.size() ? stage_variance[i+1] : true;
+            bool current_variance = stage_variance[i];
+            // if this is a single threaded stage
+            if (current_variance)
+            {
+                // 1 - 1 case
+                if (next_variance)
+                    sharedBuffers.append(new DoubleBuffer());
+                // 1 - n case
+                else
+                    sharedBuffers.append(new SingleBuffer());
+            }
+            // This is a multi-threaded stage
+            else
+            {
+                // If the next stage is single threaded, we need to sequence our
+                // output (n - 1 case)
+                if (next_variance)
+                    sharedBuffers.append(new SequencingBuffer());
+                // Otherwise, this is an n-n boundary and we don't need to
+                // adhere to any particular sequence
+                else
+                    sharedBuffers.append(new SingleBuffer());
+            }
+            processingStages.last()->outputBuffer = sharedBuffers.last();
+            processingStages.last()->transform = transforms[i];
+        }
+
+        collectionStage.inputBuffer = sharedBuffers.last();
+        collectionStage.data = &readStage.dataSource;
+        collectionStage.stage_id = next_stage_id;
+    }
+
+    ~StreamTransform()
+    {
+        for (int i = 0; i < processingStages.size(); i++) {
+            delete processingStages[i];
+        }
+        for (int i = 0; i < sharedBuffers.size(); i++) {
+            delete sharedBuffers[i];
+        }
+
+    }
+
+protected:
+    QList<bool> stage_variance;
+
+    FirstStage readStage;
+    LastStage collectionStage;
+
+    QList<ProcessingStage *> processingStages;
+    QList<SharedBuffer *> sharedBuffers;
+
+    QThreadPool processingThreads;
+
+    void _project(const Template &src, Template &dst) const
+    {
+        (void) src; (void) dst;
+        qFatal("nope");
+    }
+    void _project(const TemplateList & src, TemplateList & dst) const
+    {
+        (void) src; (void) dst;
+        qFatal("nope");
+    }
+};
+
+BR_REGISTER(Transform, StreamTransform)
+
+
+} // namespace br
+
+#include "stream.moc"
+
diff --git a/sdk/plugins/validate.cpp b/sdk/plugins/validate.cpp
index 3718ffa..4a99554 100644
--- a/sdk/plugins/validate.cpp
+++ b/sdk/plugins/validate.cpp
@@ -1,3 +1,4 @@
+#include <QFutureSynchronizer>
 #include <QtConcurrentRun>
 #include <openbr_plugin.h>
 
@@ -34,16 +35,16 @@ class CrossValidateTransform : public MetaTransform
             return;
         }
 
-        QList< QFuture<void> > futures;
+        QFutureSynchronizer<void> futures;
         for (int i=0; i<numPartitions; i++) {
             TemplateList partitionedData = data;
             for (int j=partitionedData.size()-1; j>=0; j--)
                 if (partitions[j] == i)
                     partitionedData.removeAt(j);
-            if (Globals->parallelism) futures.append(QtConcurrent::run(transforms[i], &Transform::train, partitionedData));
-            else                      transforms[i]->train(partitionedData);
+            if (Globals->parallelism) futures.addFuture(QtConcurrent::run(transforms[i], &Transform::train, partitionedData));
+            else                                                          transforms[i]->train(partitionedData);
         }
-        Globals->trackFutures(futures);
+        futures.waitForFinished();
     }
 
     void project(const Template &src, Template &dst) const
diff --git a/sdk/plugins/youtube.cpp b/sdk/plugins/youtube.cpp
index 442c4d3..6b4cca8 100644
--- a/sdk/plugins/youtube.cpp
+++ b/sdk/plugins/youtube.cpp
@@ -1,3 +1,4 @@
+#include <QCoreApplication>
 #include <QProcess>
 #include <openbr_plugin.h>
 
diff --git a/share/openbr/models b/share/openbr/models
index 5dcad73..c11ff0f 160000
--- a/share/openbr/models
+++ b/share/openbr/models
@@ -1 +1 @@
-Subproject commit 5dcad735fc16ce86937f992df8d4ba274dc81c14
+Subproject commit c11ff0f0bb451fb5576bd4b8122c32278e52042a
diff --git a/share/openbr/openbr.bib b/share/openbr/openbr.bib
index 556446d..68e6daf 100644
--- a/share/openbr/openbr.bib
+++ b/share/openbr/openbr.bib
@@ -33,7 +33,6 @@
 	Author = {Lacey S. Best-Rowden},
 	Howpublished = {https://github.com/lbestrowden},
 	Title = {bestrow1 at msu.edu}}
-	
 
 % Software
 @misc{OpenBR,
@@ -87,6 +86,14 @@
 
 
 % Papers
+@ inproceedings{Arandjelovic12, 
+	Author={Arandjelovic, R. and Zisserman, A.}, 
+	Booktitle={Computer Vision and Pattern Recognition (CVPR), 2012 IEEE Conference on},
+	Title={Three things everyone should know to improve object retrieval}, 
+	Month={June}, 
+	Year={2012}
+	Pages={2911-2918}}
+
 @inproceedings{belhumeur11,
 	Author = {Belhumeur, P.N. and Jacobs, D.W. and Kriegman, D.J. and Kumar, N.},
 	Booktitle = {Computer Vision and Pattern Recognition (CVPR), 2011 IEEE Conference on},