Moved the responsibility for automatically filling out a test_box_overlap

object from the structural_object_detection_trainer to the
structural_svm_object_detection_problem.  This allows us to use image scanners
which require an image be loaded before get_best_matching_rect() can be called.
I also made it so that the scanner loading (and therefore feature extraction)
is threaded.  Previously, it only used a single core.

dlib/svm/structural_object_detection_trainer.h

@@ -275,29 +275,8 @@ namespace dlib
             }
 #endif
 
-            test_box_overlap local_overlap_tester;
-
-            if (auto_overlap_tester)
-            {
-                std::vector<std::vector<rectangle> > mapped_rects(truth_object_detections.size());
-                for (unsigned long i = 0; i < truth_object_detections.size(); ++i)
-                {
-                    mapped_rects[i].resize(truth_object_detections[i].size());
-                    for (unsigned long j = 0; j < truth_object_detections[i].size(); ++j)
-                    {
-                        mapped_rects[i][j] = scanner.get_best_matching_rect(truth_object_detections[i][j].get_rect());
-                    }
-                }
-
-                local_overlap_tester = find_tight_overlap_tester(mapped_rects);
-            }
-            else
-            {
-                local_overlap_tester = overlap_tester;
-            }
-
             structural_svm_object_detection_problem<image_scanner_type,image_array_type > 
-                svm_prob(scanner, local_overlap_tester, images, truth_object_detections, num_threads);
+                svm_prob(scanner, overlap_tester, auto_overlap_tester, images, truth_object_detections, num_threads);
 
             if (verbose)
                 svm_prob.be_verbose();
@@ -314,7 +293,7 @@ namespace dlib
             solver(svm_prob,w);
 
             // report the results of the training.
-            return object_detector<image_scanner_type>(scanner, local_overlap_tester, w);
+            return object_detector<image_scanner_type>(scanner, svm_prob.get_overlap_tester(), w);
         }
 
         template <

dlib/svm/structural_svm_object_detection_problem.h

@@ -37,6 +37,7 @@ namespace dlib
         structural_svm_object_detection_problem(
             const image_scanner_type& scanner,
             const test_box_overlap& overlap_tester,
+            const bool auto_overlap_tester,
             const image_array_type& images_,
             const std::vector<std::vector<full_object_detection> >& truth_object_detections_,
             unsigned long num_threads = 2
@@ -75,19 +76,34 @@ namespace dlib
                 }
             }
 #endif
-
-            scanners.set_max_size(images.size());
-            scanners.set_size(images.size());
-
+            // The purpose of the max_num_dets member variable is to give us a reasonable
+            // upper limit on the number of detections we can expect from a single image.
+            // This is used in the separation_oracle to put a hard limit on the number of
+            // detections we will consider.  We do this purely for computational reasons
+            // since otherwise we can end up wasting large amounts of time on certain
+            // pathological cases during optimization which ultimately do not influence the
+            // result.  Therefore, we for the separation oracle to only consider the
+            // max_num_dets strongest detections.
             max_num_dets = 0;
             for (unsigned long i = 0; i < truth_object_detections.size(); ++i)
             {
                 if (truth_object_detections[i].size() > max_num_dets)
                     max_num_dets = truth_object_detections[i].size();
-
-                scanners[i].copy_configuration(scanner);
             }
             max_num_dets = max_num_dets*3 + 10;
+
+            initialize_scanners(scanner, num_threads);
+
+            if (auto_overlap_tester)
+            {
+                auto_configure_overlap_tester();
+            }
+        }
+
+        test_box_overlap get_overlap_tester (
+        ) const 
+        {
+            return boxes_overlap;
         }
 
         void set_match_eps (
@@ -154,6 +170,24 @@ namespace dlib
         }
 
     private:
+
+        void auto_configure_overlap_tester(
+        )
+        {
+            std::vector<std::vector<rectangle> > mapped_rects(truth_object_detections.size());
+            for (unsigned long i = 0; i < truth_object_detections.size(); ++i)
+            {
+                mapped_rects[i].resize(truth_object_detections[i].size());
+                for (unsigned long j = 0; j < truth_object_detections[i].size(); ++j)
+                {
+                    mapped_rects[i][j] = scanners[i].get_best_matching_rect(truth_object_detections[i][j].get_rect());
+                }
+            }
+
+            boxes_overlap = find_tight_overlap_tester(mapped_rects);
+        }
+
+
         virtual long get_num_dimensions (
         ) const 
         {
@@ -172,7 +206,7 @@ namespace dlib
             feature_vector_type& psi 
         ) const 
         {
-            const image_scanner_type& scanner = get_scanner(idx);
+            const image_scanner_type& scanner = scanners[idx];
 
             psi.set_size(get_num_dimensions());
             std::vector<rectangle> mapped_rects;
@@ -268,7 +302,7 @@ namespace dlib
             feature_vector_type& psi
         ) const 
         {
-            const image_scanner_type& scanner = get_scanner(idx);
+            const image_scanner_type& scanner = scanners[idx];
 
             std::vector<std::pair<double, rectangle> > dets;
             const double thresh = current_solution(scanner.get_num_dimensions());
@@ -437,13 +471,38 @@ namespace dlib
             return std::make_pair(match,best_idx);
         }
 
+        struct init_scanners_helper
+        {
+            init_scanners_helper (
+                array<image_scanner_type>& scanners_,
+                const image_array_type& images_
+            ) :
+                scanners(scanners_),
+                images(images_)
+            {}
+
+            array<image_scanner_type>& scanners;
+            const image_array_type& images;
+
+            void operator() (long i ) const
+            {
+                scanners[i].load(images[i]);
+            }
+        };
 
-        const image_scanner_type& get_scanner (long idx) const
+        void initialize_scanners (
+            const image_scanner_type& scanner,
+            unsigned long num_threads
+        )
         {
-            if (scanners[idx].is_loaded_with_image() == false)
-                scanners[idx].load(images[idx]);
+            scanners.set_max_size(images.size());
+            scanners.set_size(images.size());
+
+            for (unsigned long i = 0; i < scanners.size(); ++i)
+                scanners[i].copy_configuration(scanner);
 
-            return scanners[idx];
+            // now load the images into all the scanners
+            parallel_for(num_threads, 0, scanners.size(), init_scanners_helper(scanners, images));
         }
 
 

dlib/svm/structural_svm_object_detection_problem_abstract.h

@@ -68,7 +68,7 @@ namespace dlib
                 A detection is considered a false alarm if it doesn't match with any 
                 of the ground truth rectangles or if it is a duplicate detection of a 
                 truth rectangle.  Finally, for the purposes of calculating loss, a match 
-                is determined using the following formula, rectangles A and B match 
+                is determined using the following formula where rectangles A and B match 
                 if and only if:
                     A.intersect(B).area()/(A+B).area() > get_match_eps()
         !*/
@@ -77,6 +77,7 @@ namespace dlib
         structural_svm_object_detection_problem(
             const image_scanner_type& scanner,
             const test_box_overlap& overlap_tester,
+            const bool auto_overlap_tester,
             const image_array_type& images,
             const std::vector<std::vector<full_object_detection> >& truth_object_detections,
             unsigned long num_threads = 2
@@ -95,12 +96,18 @@ namespace dlib
                   attempts to learn to predict truth_object_detections[i] based on
                   images[i].  Or in other words, this object can be used to learn a
                   parameter vector, w, such that an object_detector declared as:
-                    object_detector<image_scanner_type> detector(scanner,overlap_tester,w)
+                    object_detector<image_scanner_type> detector(scanner,get_overlap_tester(),w)
                   results in a detector object which attempts to compute the locations of
                   all the objects in truth_object_detections.  So if you called
                   detector(images[i]) you would hopefully get a list of rectangles back
                   that had truth_object_detections[i].size() elements and contained exactly
                   the rectangles indicated by truth_object_detections[i].
+                - if (auto_overlap_tester == true) then
+                    - #get_overlap_tester() == a test_box_overlap object that is configured
+                      using the find_tight_overlap_tester() routine and the contents of
+                      truth_object_detections. 
+                - else
+                    - #get_overlap_tester() == overlap_tester
                 - #get_match_eps() == 0.5
                 - This object will use num_threads threads during the optimization 
                   procedure.  You should set this parameter equal to the number of 
@@ -109,6 +116,13 @@ namespace dlib
                 - #get_loss_per_false_alarm() == 1
         !*/
 
+        test_box_overlap get_overlap_tester (
+        ) const;
+        /*!
+            ensures
+                - returns the overlap tester used by this object.  
+        !*/
+
         void set_match_eps (
             double eps
         );