File: cross_validate_object_detection_trainer.h

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 (294 lines) | stat: -rw-r--r-- 10,080 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
// Copyright (C) 2011  Davis E. King (davis@dlib.net)
// License: Boost Software License   See LICENSE.txt for the full license.
#ifndef DLIB_CROSS_VALIDATE_OBJECT_DETECTION_TRaINER_H__
#define DLIB_CROSS_VALIDATE_OBJECT_DETECTION_TRaINER_H__

#include "cross_validate_object_detection_trainer_abstract.h"
#include <vector>
#include "../matrix.h"
#include "svm.h"
#include "../geometry.h"
#include "../image_processing/full_object_detection.h"

namespace dlib
{

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

    namespace impl
    {
        inline unsigned long number_of_truth_hits (
            const std::vector<full_object_detection>& truth_boxes,
            const std::vector<rectangle>& boxes,
            const double overlap_eps
        )
        /*!
            requires
                - 0 < overlap_eps <= 1
            ensures
                - returns the number of elements in truth_boxes which are overlapped by an 
                  element of boxes.  In this context, two boxes, A and B, overlap if and only if
                  the following quantity is greater than overlap_eps:
                    A.intersect(B).area()/(A+B).area()
                - No element of boxes is allowed to account for more than one element of truth_boxes.  
                - The returned number is in the range [0,truth_boxes.size()]
        !*/
        {
            if (boxes.size() == 0)
                return 0;

            unsigned long count = 0;
            std::vector<bool> used(boxes.size(),false);
            for (unsigned long i = 0; i < truth_boxes.size(); ++i)
            {
                unsigned long best_idx = 0;
                double best_overlap = 0;
                for (unsigned long j = 0; j < boxes.size(); ++j)
                {
                    if (used[j])
                        continue;

                    const double overlap = truth_boxes[i].get_rect().intersect(boxes[j]).area() / (double)(truth_boxes[i].get_rect()+boxes[j]).area();
                    if (overlap > best_overlap)
                    {
                        best_overlap = overlap;
                        best_idx = j;
                    }
                }

                if (best_overlap > overlap_eps && used[best_idx] == false)
                {
                    used[best_idx] = true;
                    ++count;
                }
            }

            return count;
        }
    }

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

    template <
        typename object_detector_type,
        typename image_array_type
        >
    const matrix<double,1,2> test_object_detection_function (
        object_detector_type& detector,
        const image_array_type& images,
        const std::vector<std::vector<full_object_detection> >& truth_dets,
        const double overlap_eps = 0.5
    )
    {
        // make sure requires clause is not broken
        DLIB_ASSERT( is_learning_problem(images,truth_dets) == true &&
                     0 < overlap_eps && overlap_eps <= 1,
                    "\t matrix test_object_detection_function()"
                    << "\n\t invalid inputs were given to this function"
                    << "\n\t is_learning_problem(images,truth_dets): " << is_learning_problem(images,truth_dets)
                    << "\n\t overlap_eps: "<< overlap_eps
                    );



        double correct_hits = 0;
        double total_hits = 0;
        double total_true_targets = 0;

        for (unsigned long i = 0; i < images.size(); ++i)
        {
            const std::vector<rectangle>& hits = detector(images[i]);

            total_hits += hits.size();
            correct_hits += impl::number_of_truth_hits(truth_dets[i], hits, overlap_eps);
            total_true_targets += truth_dets[i].size();
        }


        double precision, recall;

        if (total_hits == 0)
            precision = 1;
        else
            precision = correct_hits / total_hits;

        if (total_true_targets == 0)
            recall = 1;
        else
            recall = correct_hits / total_true_targets;

        matrix<double, 1, 2> res;
        res = precision, recall;
        return res;
    }

    template <
        typename object_detector_type,
        typename image_array_type
        >
    const matrix<double,1,2> test_object_detection_function (
        object_detector_type& detector,
        const image_array_type& images,
        const std::vector<std::vector<rectangle> >& truth_dets,
        const double overlap_eps = 0.5
    )
    {
        // convert into a list of regular rectangles.
        std::vector<std::vector<full_object_detection> > rects(truth_dets.size());
        for (unsigned long i = 0; i < truth_dets.size(); ++i)
        {
            for (unsigned long j = 0; j < truth_dets[i].size(); ++j)
            {
                rects[i].push_back(full_object_detection(truth_dets[i][j]));
            }
        }

        return test_object_detection_function(detector, images, rects, overlap_eps);
    }

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

    namespace impl
    {
        template <
            typename array_type
            >
        struct array_subset_helper
        {
            typedef typename array_type::mem_manager_type mem_manager_type;

            array_subset_helper (
                const array_type& array_,
                const std::vector<unsigned long>& idx_set_
            ) :
                array(array_),
                idx_set(idx_set_)
            {
            }

            unsigned long size() const { return idx_set.size(); }

            typedef typename array_type::type type;
            const type& operator[] (
                unsigned long idx
            ) const { return array[idx_set[idx]]; }

        private:
            const array_type& array;
            const std::vector<unsigned long>& idx_set;
        };
    }

// ----------------------------------------------------------------------------------------
    
    template <
        typename trainer_type,
        typename image_array_type
        >
    const matrix<double,1,2> cross_validate_object_detection_trainer (
        const trainer_type& trainer,
        const image_array_type& images,
        const std::vector<std::vector<full_object_detection> >& truth_dets,
        const long folds,
        const double overlap_eps = 0.5
    )
    {
        // make sure requires clause is not broken
        DLIB_ASSERT( is_learning_problem(images,truth_dets) == true &&
                     0 < overlap_eps && overlap_eps <= 1 &&
                     1 < folds && folds <= static_cast<long>(images.size()),
                    "\t matrix cross_validate_object_detection_trainer()"
                    << "\n\t invalid inputs were given to this function"
                    << "\n\t is_learning_problem(images,truth_dets): " << is_learning_problem(images,truth_dets)
                    << "\n\t overlap_eps: "<< overlap_eps
                    << "\n\t folds: "<< folds
                    );

        double correct_hits = 0;
        double total_hits = 0;
        double total_true_targets = 0;

        const long test_size = images.size()/folds;

        unsigned long test_idx = 0;
        for (long iter = 0; iter < folds; ++iter)
        {
            std::vector<unsigned long> train_idx_set;
            std::vector<unsigned long> test_idx_set;

            for (long i = 0; i < test_size; ++i)
                test_idx_set.push_back(test_idx++);

            unsigned long train_idx = test_idx%images.size();
            std::vector<std::vector<full_object_detection> > training_rects;
            for (unsigned long i = 0; i < images.size()-test_size; ++i)
            {
                training_rects.push_back(truth_dets[train_idx]);
                train_idx_set.push_back(train_idx);
                train_idx = (train_idx+1)%images.size();
            }


            impl::array_subset_helper<image_array_type> array_subset(images, train_idx_set);
            typename trainer_type::trained_function_type detector = trainer.train(array_subset, training_rects);
            for (unsigned long i = 0; i < test_idx_set.size(); ++i)
            {
                const std::vector<rectangle>& hits = detector(images[test_idx_set[i]]);

                total_hits += hits.size();
                correct_hits += impl::number_of_truth_hits(truth_dets[test_idx_set[i]], hits, overlap_eps);
                total_true_targets += truth_dets[test_idx_set[i]].size();
            }

        }



        double precision, recall;

        if (total_hits == 0)
            precision = 1;
        else
            precision = correct_hits / total_hits;

        if (total_true_targets == 0)
            recall = 1;
        else
            recall = correct_hits / total_true_targets;

        matrix<double, 1, 2> res;
        res = precision, recall;
        return res;
    }

    template <
        typename trainer_type,
        typename image_array_type
        >
    const matrix<double,1,2> cross_validate_object_detection_trainer (
        const trainer_type& trainer,
        const image_array_type& images,
        const std::vector<std::vector<rectangle> >& truth_dets,
        const long folds,
        const double overlap_eps = 0.5
    )
    {
        // convert into a list of regular rectangles.
        std::vector<std::vector<full_object_detection> > dets(truth_dets.size());
        for (unsigned long i = 0; i < truth_dets.size(); ++i)
        {
            for (unsigned long j = 0; j < truth_dets[i].size(); ++j)
            {
                dets[i].push_back(full_object_detection(truth_dets[i][j]));
            }
        }

        return cross_validate_object_detection_trainer(trainer, images, dets, folds, overlap_eps);
    }

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

}

#endif // DLIB_CROSS_VALIDATE_OBJECT_DETECTION_TRaINER_H__