File: object_detector.cpp

package info (click to toggle)
mldemos 0.5.1-3
  • links: PTS, VCS
  • area: main
  • in suites: jessie, jessie-kfreebsd
  • size: 32,224 kB
  • ctags: 46,525
  • sloc: cpp: 306,887; ansic: 167,718; ml: 126; sh: 109; makefile: 2
file content (779 lines) | stat: -rw-r--r-- 31,798 bytes parent folder | download | duplicates (2)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
// Copyright (C) 2011  Davis E. King (davis@dlib.net)
// License: Boost Software License   See LICENSE.txt for the full license.


#include <dlib/statistics.h>
#include <sstream>
#include <string>
#include <cstdlib>
#include <ctime>
#include "tester.h"
#include <dlib/pixel.h>
#include <dlib/svm_threaded.h>
#include <dlib/array.h>
#include <dlib/array2d.h>
#include <dlib/image_keypoint.h>
#include <dlib/image_processing.h>
#include <dlib/image_transforms.h>

namespace  
{
    using namespace test;
    using namespace dlib;
    using namespace std;

    logger dlog("test.object_detector");

// ----------------------------------------------------------------------------------------

    struct funny_image
    {
        array2d<unsigned char> img;
        long nr() const { return img.nr(); }
        long nc() const { return img.nc(); }
    };

    void swap(funny_image& a, funny_image& b)
    {
        a.img.swap(b.img);
    }

// ----------------------------------------------------------------------------------------

    template <
        typename image_array_type,
        typename detector_type
        >
    void validate_some_object_detector_stuff (
        const image_array_type& images,
        detector_type& detector
    )
    {
        for (unsigned long i = 0; i < images.size(); ++i)
        {
            std::vector<rectangle> dets = detector(images[i]);
            std::vector<std::pair<double,rectangle> > dets2;

            detector(images[i], dets2);

            matrix<double,0,1> psi(detector.get_w().size());
            matrix<double,0,1> psi2(detector.get_w().size());
            const double thresh = detector.get_w()(detector.get_w().size()-1);

            DLIB_TEST(dets.size() == dets2.size());
            for (unsigned long j = 0; j < dets.size(); ++j)
            {
                DLIB_TEST(dets[j] == dets2[j].second);

                const full_object_detection fdet = detector.get_scanner().get_full_object_detection(dets[j], detector.get_w());
                psi = 0;
                detector.get_scanner().get_feature_vector(fdet, psi);

                double check_score = dot(psi,detector.get_w()) - thresh;
                DLIB_TEST(std::abs(check_score - dets2[j].first) < 1e-10);
            }

        }
    }

// ----------------------------------------------------------------------------------------

    class very_simple_feature_extractor : noncopyable
    {
        /*!
        WHAT THIS OBJECT REPRESENTS
            This object is a feature extractor which goes to every pixel in an image and
            produces a 32 dimensional feature vector.  This vector is an indicator vector
            which records the pattern of pixel values in a 4-connected region.  So it should
            be able to distinguish basic things like whether or not a location falls on the
            corner of a white box, on an edge, in the middle, etc.


            Note that this object also implements the interface defined in dlib/image_keypoint/hashed_feature_image_abstract.h.
            This means all the member functions in this object are supposed to behave as 
            described in the hashed_feature_image specification.  So when you define your own
            feature extractor objects you should probably refer yourself to that documentation
            in addition to reading this example program.
        !*/


    public:

        inline void load (
            const funny_image& img_
        )
        {
            const array2d<unsigned char>& img = img_.img;

            feat_image.set_size(img.nr(), img.nc());
            assign_all_pixels(feat_image,0);
            for (long r = 1; r+1 < img.nr(); ++r)
            {
                for (long c = 1; c+1 < img.nc(); ++c)
                {
                    unsigned char f = 0;
                    if (img[r][c])   f |= 0x1;
                    if (img[r][c+1]) f |= 0x2;
                    if (img[r][c-1]) f |= 0x4;
                    if (img[r+1][c]) f |= 0x8;
                    if (img[r-1][c]) f |= 0x10;

                    // Store the code value for the pattern of pixel values in the 4-connected
                    // neighborhood around this row and column.
                    feat_image[r][c] = f;
                }
            }
        }

        inline void load (
            const array2d<unsigned char>& img
        )
        {
            feat_image.set_size(img.nr(), img.nc());
            assign_all_pixels(feat_image,0);
            for (long r = 1; r+1 < img.nr(); ++r)
            {
                for (long c = 1; c+1 < img.nc(); ++c)
                {
                    unsigned char f = 0;
                    if (img[r][c])   f |= 0x1;
                    if (img[r][c+1]) f |= 0x2;
                    if (img[r][c-1]) f |= 0x4;
                    if (img[r+1][c]) f |= 0x8;
                    if (img[r-1][c]) f |= 0x10;

                    // Store the code value for the pattern of pixel values in the 4-connected
                    // neighborhood around this row and column.
                    feat_image[r][c] = f;
                }
            }
        }

        inline unsigned long size () const { return feat_image.size(); }
        inline long nr () const { return feat_image.nr(); }
        inline long nc () const { return feat_image.nc(); }

        inline long get_num_dimensions (
        ) const
        {
            // Return the dimensionality of the vectors produced by operator()
            return 32;
        }

        typedef std::vector<std::pair<unsigned int,double> > descriptor_type;

        inline const descriptor_type& operator() (
            long row,
            long col
        ) const
            /*!
                requires
                    - 0 <= row < nr()
            - 0 <= col < nc()
                ensures
                    - returns a sparse vector which describes the image at the given row and column.  
                      In particular, this is a vector that is 0 everywhere except for one element. 
            !*/
        {
            feat.clear();
            const unsigned long only_nonzero_element_index = feat_image[row][col];
            feat.push_back(make_pair(only_nonzero_element_index,1.0));
            return feat;
        }

        // This block of functions is meant to provide a way to map between the row/col space taken by
        // this object's operator() function and the images supplied to load().  In this example it's trivial.  
        // However, in general, you might create feature extractors which don't perform extraction at every 
        // possible image location (e.g. the hog_image) and thus result in some more complex mapping.  
        inline const rectangle get_block_rect       ( long row, long col) const { return centered_rect(col,row,3,3); }
        inline const point image_to_feat_space      ( const point& p) const { return p; } 
        inline const rectangle image_to_feat_space  ( const rectangle& rect) const { return rect; } 
        inline const point feat_to_image_space      ( const point& p) const { return p; } 
        inline const rectangle feat_to_image_space  ( const rectangle& rect) const { return rect; }

        inline friend void serialize   ( const very_simple_feature_extractor& item, std::ostream& out)  { serialize(item.feat_image, out); }
        inline friend void deserialize ( very_simple_feature_extractor& item, std::istream& in ) { deserialize(item.feat_image, in); }

        void copy_configuration ( const very_simple_feature_extractor& ){}

    private:
        array2d<unsigned char> feat_image;

        // This variable doesn't logically contribute to the state of this object.  It is here
        // only to avoid returning a descriptor_type object by value inside the operator() method.
        mutable descriptor_type feat;
    };

// ----------------------------------------------------------------------------------------

    template <
        typename image_array_type
        >
    void make_simple_test_data (
        image_array_type& images,
        std::vector<std::vector<rectangle> >& object_locations
    )
    {
        images.clear();
        object_locations.clear();

        images.resize(3);
        images[0].set_size(400,400);
        images[1].set_size(400,400);
        images[2].set_size(400,400);

        // set all the pixel values to black
        assign_all_pixels(images[0], 0);
        assign_all_pixels(images[1], 0);
        assign_all_pixels(images[2], 0);

        // Now make some squares and draw them onto our black images. All the
        // squares will be 70 pixels wide and tall.

        std::vector<rectangle> temp;
        temp.push_back(centered_rect(point(100,100), 70,70)); 
        fill_rect(images[0],temp.back(),255); // Paint the square white
        temp.push_back(centered_rect(point(200,300), 70,70));
        fill_rect(images[0],temp.back(),255); // Paint the square white
        object_locations.push_back(temp);

        temp.clear();
        temp.push_back(centered_rect(point(140,200), 70,70));
        fill_rect(images[1],temp.back(),255); // Paint the square white
        temp.push_back(centered_rect(point(303,200), 70,70));
        fill_rect(images[1],temp.back(),255); // Paint the square white
        object_locations.push_back(temp);

        temp.clear();
        temp.push_back(centered_rect(point(123,121), 70,70));
        fill_rect(images[2],temp.back(),255); // Paint the square white
        object_locations.push_back(temp);
    }

    template <
        typename image_array_type
        >
    void make_simple_test_data (
        image_array_type& images,
        std::vector<std::vector<full_object_detection> >& object_locations
    )
    {
        images.clear();
        object_locations.clear();


        images.resize(3);
        images[0].set_size(400,400);
        images[1].set_size(400,400);
        images[2].set_size(400,400);

        // set all the pixel values to black
        assign_all_pixels(images[0], 0);
        assign_all_pixels(images[1], 0);
        assign_all_pixels(images[2], 0);

        // Now make some squares and draw them onto our black images. All the
        // squares will be 70 pixels wide and tall.
        const int shrink = 0;
        std::vector<full_object_detection> temp;

        rectangle rect = centered_rect(point(100,100), 70,71);
        std::vector<point> movable_parts;
        movable_parts.push_back(shrink_rect(rect,shrink).tl_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).tr_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).bl_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).br_corner());
        temp.push_back(full_object_detection(rect, movable_parts)); 
        fill_rect(images[0],rect,255); // Paint the square white

        rect = centered_rect(point(200,200), 70,71);
        movable_parts.clear();
        movable_parts.push_back(shrink_rect(rect,shrink).tl_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).tr_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).bl_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).br_corner());
        temp.push_back(full_object_detection(rect, movable_parts)); 
        fill_rect(images[0],rect,255); // Paint the square white

        object_locations.push_back(temp);
        // ------------------------------------
        temp.clear();

        rect = centered_rect(point(140,200), 70,71);
        movable_parts.clear();
        movable_parts.push_back(shrink_rect(rect,shrink).tl_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).tr_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).bl_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).br_corner());
        temp.push_back(full_object_detection(rect, movable_parts)); 
        fill_rect(images[1],rect,255); // Paint the square white


        rect = centered_rect(point(303,200), 70,71);
        movable_parts.clear();
        movable_parts.push_back(shrink_rect(rect,shrink).tl_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).tr_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).bl_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).br_corner());
        temp.push_back(full_object_detection(rect, movable_parts)); 
        fill_rect(images[1],rect,255); // Paint the square white

        object_locations.push_back(temp);
        // ------------------------------------
        temp.clear();

        rect = centered_rect(point(123,121), 70,71);
        movable_parts.clear();
        movable_parts.push_back(shrink_rect(rect,shrink).tl_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).tr_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).bl_corner());
        movable_parts.push_back(shrink_rect(rect,shrink).br_corner());
        temp.push_back(full_object_detection(rect, movable_parts)); 
        fill_rect(images[2],rect,255); // Paint the square white

        object_locations.push_back(temp);

        // corrupt each image with random noise just to make this a little more 
        // challenging 
        dlib::rand rnd;
        for (unsigned long i = 0; i < images.size(); ++i)
        {
            for (long r = 0; r < images[i].nr(); ++r)
            {
                for (long c = 0; c < images[i].nc(); ++c)
                {
                    images[i][r][c] = put_in_range(0,255,images[i][r][c] + 40*rnd.get_random_gaussian());
                }
            }
        }
    }
// ----------------------------------------------------------------------------------------

    void test_1 (
    )
    {        
        print_spinner();
        dlog << LINFO << "test_1()";

        typedef array<array2d<unsigned char> >  grayscale_image_array_type;
        grayscale_image_array_type images;
        std::vector<std::vector<rectangle> > object_locations;
        make_simple_test_data(images, object_locations);

        typedef hashed_feature_image<hog_image<3,3,1,4,hog_signed_gradient,hog_full_interpolation> > feature_extractor_type;
        typedef scan_image_pyramid<pyramid_down, feature_extractor_type> image_scanner_type;
        image_scanner_type scanner;
        const rectangle object_box = compute_box_dimensions(1,35*35);
        scanner.add_detection_template(object_box, create_grid_detection_template(object_box,2,2));
        setup_hashed_features(scanner, images, 9);
        structural_object_detection_trainer<image_scanner_type> trainer(scanner);
        trainer.set_num_threads(4);  
        trainer.set_overlap_tester(test_box_overlap(0,0));
        object_detector<image_scanner_type> detector = trainer.train(images, object_locations);

        matrix<double> res = test_object_detection_function(detector, images, object_locations);
        dlog << LINFO << "Test detector (precision,recall): " << res;
        DLIB_TEST(sum(res) == 2);

        {
            ostringstream sout;
            serialize(detector, sout);
            istringstream sin(sout.str());
            object_detector<image_scanner_type> d2;
            deserialize(d2, sin);
            matrix<double> res = test_object_detection_function(d2, images, object_locations);
            dlog << LINFO << "Test detector (precision,recall): " << res;
            DLIB_TEST(sum(res) == 2);

            validate_some_object_detector_stuff(images, detector);
        }
    }

// ----------------------------------------------------------------------------------------

    void test_1m (
    )
    {        
        print_spinner();
        dlog << LINFO << "test_1m()";

        typedef array<array2d<unsigned char> >  grayscale_image_array_type;
        grayscale_image_array_type images;
        std::vector<std::vector<full_object_detection> > object_locations;
        make_simple_test_data(images, object_locations);

        typedef hashed_feature_image<hog_image<3,3,1,4,hog_signed_gradient,hog_full_interpolation> > feature_extractor_type;
        typedef scan_image_pyramid<pyramid_down, feature_extractor_type> image_scanner_type;
        image_scanner_type scanner;
        const rectangle object_box = compute_box_dimensions(1,35*35);
        std::vector<rectangle> mboxes;
        const int mbox_size = 20;
        mboxes.push_back(centered_rect(0,0, mbox_size,mbox_size));
        mboxes.push_back(centered_rect(0,0, mbox_size,mbox_size));
        mboxes.push_back(centered_rect(0,0, mbox_size,mbox_size));
        mboxes.push_back(centered_rect(0,0, mbox_size,mbox_size));
        scanner.add_detection_template(object_box, create_grid_detection_template(object_box,1,1), mboxes);
        setup_hashed_features(scanner, images, 9);
        structural_object_detection_trainer<image_scanner_type> trainer(scanner);
        trainer.set_num_threads(4);  
        trainer.set_overlap_tester(test_box_overlap(0,0));
        object_detector<image_scanner_type> detector = trainer.train(images, object_locations);

        matrix<double> res = test_object_detection_function(detector, images, object_locations);
        dlog << LINFO << "Test detector (precision,recall): " << res;
        DLIB_TEST(sum(res) == 2);

        {
            ostringstream sout;
            serialize(detector, sout);
            istringstream sin(sout.str());
            object_detector<image_scanner_type> d2;
            deserialize(d2, sin);
            matrix<double> res = test_object_detection_function(d2, images, object_locations);
            dlog << LINFO << "Test detector (precision,recall): " << res;
            DLIB_TEST(sum(res) == 2);

            validate_some_object_detector_stuff(images, detector);
        }
    }

// ----------------------------------------------------------------------------------------

    void test_1_fine_hog (
    )
    {        
        print_spinner();
        dlog << LINFO << "test_1_fine_hog()";

        typedef array<array2d<unsigned char> >  grayscale_image_array_type;
        grayscale_image_array_type images;
        std::vector<std::vector<rectangle> > object_locations;
        make_simple_test_data(images, object_locations);

        typedef hashed_feature_image<fine_hog_image<3,3,2,4,hog_signed_gradient> > feature_extractor_type;
        typedef scan_image_pyramid<pyramid_down, feature_extractor_type> image_scanner_type;
        image_scanner_type scanner;
        const rectangle object_box = compute_box_dimensions(1,35*35);
        scanner.add_detection_template(object_box, create_grid_detection_template(object_box,2,2));
        setup_hashed_features(scanner, images, 9);
        structural_object_detection_trainer<image_scanner_type> trainer(scanner);
        trainer.set_num_threads(4);  
        trainer.set_overlap_tester(test_box_overlap(0,0));
        object_detector<image_scanner_type> detector = trainer.train(images, object_locations);

        matrix<double> res = test_object_detection_function(detector, images, object_locations);
        dlog << LINFO << "Test detector (precision,recall): " << res;
        DLIB_TEST(sum(res) == 2);

        {
            ostringstream sout;
            serialize(detector, sout);
            istringstream sin(sout.str());
            object_detector<image_scanner_type> d2;
            deserialize(d2, sin);
            matrix<double> res = test_object_detection_function(d2, images, object_locations);
            dlog << LINFO << "Test detector (precision,recall): " << res;
            DLIB_TEST(sum(res) == 2);

            validate_some_object_detector_stuff(images, detector);
        }
    }

// ----------------------------------------------------------------------------------------

    void test_1_poly (
    )
    {        
        print_spinner();
        dlog << LINFO << "test_1_poly()";

        typedef array<array2d<unsigned char> >  grayscale_image_array_type;
        grayscale_image_array_type images;
        std::vector<std::vector<rectangle> > object_locations;
        make_simple_test_data(images, object_locations);

        typedef hashed_feature_image<poly_image<2> > feature_extractor_type;
        typedef scan_image_pyramid<pyramid_down, feature_extractor_type> image_scanner_type;
        image_scanner_type scanner;
        const rectangle object_box = compute_box_dimensions(1,35*35);
        scanner.add_detection_template(object_box, create_grid_detection_template(object_box,2,2));
        setup_hashed_features(scanner, images, 9);
        structural_object_detection_trainer<image_scanner_type> trainer(scanner);
        trainer.set_num_threads(4);  
        trainer.set_overlap_tester(test_box_overlap(0,0));
        object_detector<image_scanner_type> detector = trainer.train(images, object_locations);

        matrix<double> res = test_object_detection_function(detector, images, object_locations);
        dlog << LINFO << "Test detector (precision,recall): " << res;
        DLIB_TEST(sum(res) == 2);

        {
            ostringstream sout;
            serialize(detector, sout);
            istringstream sin(sout.str());
            object_detector<image_scanner_type> d2;
            deserialize(d2, sin);
            matrix<double> res = test_object_detection_function(d2, images, object_locations);
            dlog << LINFO << "Test detector (precision,recall): " << res;
            DLIB_TEST(sum(res) == 2);

            validate_some_object_detector_stuff(images, detector);
        }
    }

// ----------------------------------------------------------------------------------------

    void test_1m_poly (
    )
    {        
        print_spinner();
        dlog << LINFO << "test_1_poly()";

        typedef array<array2d<unsigned char> >  grayscale_image_array_type;
        grayscale_image_array_type images;
        std::vector<std::vector<full_object_detection> > object_locations;
        make_simple_test_data(images, object_locations);

        typedef hashed_feature_image<poly_image<2> > feature_extractor_type;
        typedef scan_image_pyramid<pyramid_down_3_2, feature_extractor_type> image_scanner_type;
        image_scanner_type scanner;
        const rectangle object_box = compute_box_dimensions(1,35*35);
        std::vector<rectangle> mboxes;
        const int mbox_size = 20;
        mboxes.push_back(centered_rect(0,0, mbox_size,mbox_size));
        mboxes.push_back(centered_rect(0,0, mbox_size,mbox_size));
        mboxes.push_back(centered_rect(0,0, mbox_size,mbox_size));
        mboxes.push_back(centered_rect(0,0, mbox_size,mbox_size));
        scanner.add_detection_template(object_box, create_grid_detection_template(object_box,2,2), mboxes);
        setup_hashed_features(scanner, images, 9);
        structural_object_detection_trainer<image_scanner_type> trainer(scanner);
        trainer.set_num_threads(4);  
        trainer.set_overlap_tester(test_box_overlap(0,0));
        object_detector<image_scanner_type> detector = trainer.train(images, object_locations);

        matrix<double> res = test_object_detection_function(detector, images, object_locations);
        dlog << LINFO << "Test detector (precision,recall): " << res;
        DLIB_TEST(sum(res) == 2);

        {
            ostringstream sout;
            serialize(detector, sout);
            istringstream sin(sout.str());
            object_detector<image_scanner_type> d2;
            deserialize(d2, sin);
            matrix<double> res = test_object_detection_function(d2, images, object_locations);
            dlog << LINFO << "Test detector (precision,recall): " << res;
            DLIB_TEST(sum(res) == 2);

            validate_some_object_detector_stuff(images, detector);
        }
    }

// ----------------------------------------------------------------------------------------

    void test_1_poly_nn (
    )
    {        
        print_spinner();
        dlog << LINFO << "test_1_poly_nn()";

        typedef array<array2d<unsigned char> >  grayscale_image_array_type;
        grayscale_image_array_type images;
        std::vector<std::vector<rectangle> > object_locations;
        make_simple_test_data(images, object_locations);

        typedef nearest_neighbor_feature_image<poly_image<5> > feature_extractor_type;
        typedef scan_image_pyramid<pyramid_down, feature_extractor_type> image_scanner_type;
        image_scanner_type scanner;

        setup_grid_detection_templates(scanner, object_locations, 2, 2);
        feature_extractor_type nnfe;
        pyramid_down pyr_down;
        poly_image<5> polyi;
        nnfe.set_basis(randomly_sample_image_features(images, pyr_down, polyi, 80));
        scanner.copy_configuration(nnfe);

        structural_object_detection_trainer<image_scanner_type> trainer(scanner);
        trainer.set_num_threads(4);  
        object_detector<image_scanner_type> detector = trainer.train(images, object_locations);

        matrix<double> res = test_object_detection_function(detector, images, object_locations);
        dlog << LINFO << "Test detector (precision,recall): " << res;
        DLIB_TEST(sum(res) == 2);

        {
            ostringstream sout;
            serialize(detector, sout);
            istringstream sin(sout.str());
            object_detector<image_scanner_type> d2;
            deserialize(d2, sin);
            matrix<double> res = test_object_detection_function(d2, images, object_locations);
            dlog << LINFO << "Test detector (precision,recall): " << res;
            DLIB_TEST(sum(res) == 2);

            validate_some_object_detector_stuff(images, detector);
        }
    }

// ----------------------------------------------------------------------------------------

    void test_2 (
    )
    {        
        print_spinner();
        dlog << LINFO << "test_2()";

        typedef array<array2d<unsigned char> >  grayscale_image_array_type;
        grayscale_image_array_type images;
        std::vector<std::vector<rectangle> > object_locations;
        make_simple_test_data(images, object_locations);

        typedef scan_image_pyramid<pyramid_down_5_4, very_simple_feature_extractor> image_scanner_type;
        image_scanner_type scanner;
        const rectangle object_box = compute_box_dimensions(1,70*70);
        scanner.add_detection_template(object_box, create_grid_detection_template(object_box,2,2));
        scanner.set_max_pyramid_levels(1);
        structural_object_detection_trainer<image_scanner_type> trainer(scanner);
        trainer.set_num_threads(0);  
        object_detector<image_scanner_type> detector = trainer.train(images, object_locations);

        matrix<double> res = test_object_detection_function(detector, images, object_locations);
        dlog << LINFO << "Test detector (precision,recall): " << res;
        DLIB_TEST(sum(res) == 2);

        res = cross_validate_object_detection_trainer(trainer, images, object_locations, 3);
        dlog << LINFO << "3-fold cross validation (precision,recall): " << res;
        DLIB_TEST(sum(res) == 2);

        {
            ostringstream sout;
            serialize(detector, sout);
            istringstream sin(sout.str());
            object_detector<image_scanner_type> d2;
            deserialize(d2, sin);
            matrix<double> res = test_object_detection_function(d2, images, object_locations);
            dlog << LINFO << "Test detector (precision,recall): " << res;
            DLIB_TEST(sum(res) == 2);
            validate_some_object_detector_stuff(images, detector);
        }
    }

// ----------------------------------------------------------------------------------------

    class pyramid_down_funny : noncopyable
    {
        pyramid_down pyr;
    public:

        template <typename T>
        dlib::vector<double,2> point_down ( const dlib::vector<T,2>& p) const { return pyr.point_down(p); }

        template <typename T>
        dlib::vector<double,2> point_up ( const dlib::vector<T,2>& p) const { return pyr.point_up(p); }

        template <typename T>
        dlib::vector<double,2> point_down ( const dlib::vector<T,2>& p, unsigned int levels) const { return pyr.point_down(p,levels); }

        template <typename T>
        dlib::vector<double,2> point_up ( const dlib::vector<T,2>& p, unsigned int levels) const { return pyr.point_up(p,levels); }

        rectangle rect_up ( const rectangle& rect) const { return pyr.rect_up(rect); }

        rectangle rect_up ( const rectangle& rect, unsigned int levels) const { return pyr.rect_up(rect,levels); }

        rectangle rect_down ( const rectangle& rect) const { return pyr.rect_down(rect); }

        rectangle rect_down ( const rectangle& rect, unsigned int levels) const { return pyr.rect_down(rect,levels); }

        template <
            typename in_image_type,
            typename out_image_type
            >
        void operator() (
            const in_image_type& original,
            out_image_type& down
        ) const
        {
            pyr(original.img, down.img);
        }

    };

    // make sure everything works even when the image isn't a dlib::array2d.
    // So test with funny_image.
    void test_3 (
    )
    {        
        print_spinner();
        dlog << LINFO << "test_3()";


        typedef array<array2d<unsigned char> >  grayscale_image_array_type;
        typedef array<funny_image>  funny_image_array_type;
        grayscale_image_array_type images_temp;
        funny_image_array_type images;
        std::vector<std::vector<rectangle> > object_locations;
        make_simple_test_data(images_temp, object_locations);
        images.resize(images_temp.size());
        for (unsigned long i = 0; i < images_temp.size(); ++i)
        {
            images[i].img.swap(images_temp[i]);
        }

        typedef scan_image_pyramid<pyramid_down_funny, very_simple_feature_extractor> image_scanner_type;
        image_scanner_type scanner;
        const rectangle object_box = compute_box_dimensions(1,70*70);
        scanner.add_detection_template(object_box, create_grid_detection_template(object_box,2,2));
        scanner.set_max_pyramid_levels(1);
        structural_object_detection_trainer<image_scanner_type> trainer(scanner);
        trainer.set_num_threads(4);  
        object_detector<image_scanner_type> detector = trainer.train(images, object_locations);

        matrix<double> res = test_object_detection_function(detector, images, object_locations);
        dlog << LINFO << "Test detector (precision,recall): " << res;
        DLIB_TEST(sum(res) == 2);

        res = cross_validate_object_detection_trainer(trainer, images, object_locations, 3);
        dlog << LINFO << "3-fold cross validation (precision,recall): " << res;
        DLIB_TEST(sum(res) == 2);

        {
            ostringstream sout;
            serialize(detector, sout);
            istringstream sin(sout.str());
            object_detector<image_scanner_type> d2;
            deserialize(d2, sin);
            matrix<double> res = test_object_detection_function(d2, images, object_locations);
            dlog << LINFO << "Test detector (precision,recall): " << res;
            DLIB_TEST(sum(res) == 2);
        }
    }

// ----------------------------------------------------------------------------------------

    class object_detector_tester : public tester
    {
    public:
        object_detector_tester (
        ) :
            tester ("test_object_detector",
                    "Runs tests on the structural object detection stuff.")
        {}

        void perform_test (
        )
        {
            test_1();
            test_1m();
            test_1_fine_hog();
            test_1_poly();
            test_1m_poly();
            test_1_poly_nn();
            test_2();
            test_3();
        }
    } a;

}