dlib.cpp
6.06 KB
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
#include <opencv2/imgproc/imgproc.hpp>
#include <dlib/image_processing/frontal_face_detector.h>
#include <dlib/opencv.h>
#include "openbr/plugins/openbr_internal.h"
#include <QTemporaryFile>
using namespace std;
using namespace dlib;
namespace br
{
/*!
* \ingroup transforms
* \brief Wrapper to dlib's landmarker.
* \author Scott Klum \cite sklum
*/
class DLibShapeResourceMaker : public ResourceMaker<shape_predictor>
{
private:
shape_predictor *make() const
{
shape_predictor *sp = new shape_predictor();
dlib::deserialize(qPrintable(Globals->sdkPath + "/share/openbr/models/dlib/shape_predictor_68_face_landmarks.dat")) >> *sp;
return sp;
}
};
class DLandmarkerTransform : public UntrainableTransform
{
Q_OBJECT
private:
Resource<shape_predictor> shapeResource;
void init()
{
shapeResource.setResourceMaker(new DLibShapeResourceMaker());
}
QPointF averagePoints(const QList<QPointF> &points, int rangeBegin, int rangeEnd) const
{
QPointF point;
for (int i=rangeBegin; i<rangeEnd; i++)
point += points[i];
point /= (rangeEnd-rangeBegin);
return point;
}
void setFacePoints(Template &dst) const
{
const QList<QPointF> points = dst.file.points();
dst.file.set("LeftEye",averagePoints(points,36,42));
dst.file.set("RightEye",averagePoints(points,42,48));
dst.file.set("Chin",points[8]);
}
void project(const Template &src, Template &dst) const
{
dst = src;
shape_predictor *const sp = shapeResource.acquire();
cv::Mat cvImage = src.m();
if (cvImage.channels() == 3)
cv::cvtColor(cvImage, cvImage, CV_BGR2GRAY);
cv_image<unsigned char> cimg(cvImage);
array2d<unsigned char> image;
assign_image(image,cimg);
if (src.file.rects().isEmpty()) { //If the image has no rects assume the whole image is a face
rectangle r(0, 0, cvImage.cols, cvImage.rows);
full_object_detection shape = (*sp)(image, r);
for (size_t i = 0; i < shape.num_parts(); i++)
dst.file.appendPoint(QPointF(shape.part(i)(0),shape.part(i)(1)));
setFacePoints(dst);
}
else { // Crop the image on the rects
for (int j=0; j<src.file.rects().size(); ++j)
{
QRectF rect = src.file.rects()[j];
rectangle r(rect.left(),rect.top(),rect.right(),rect.bottom());
full_object_detection shape = (*sp)(image, r);
for (size_t i=0; i<shape.num_parts(); i++)
dst.file.appendPoint(QPointF(shape.part(i)(0),shape.part(i)(1)));
setFacePoints(dst);
}
}
shapeResource.release(sp);
}
};
BR_REGISTER(Transform, DLandmarkerTransform)
/*!
* \ingroup transforms
* \brief Wrapper to dlib's trainable object detector.
* \author Scott Klum \cite sklum
*/
class DObjectDetectorTransform : public Transform
{
Q_OBJECT
Q_PROPERTY(int winSize READ get_winSize WRITE set_winSize RESET reset_winSize STORED true)
Q_PROPERTY(float C READ get_C WRITE set_C RESET reset_C STORED true)
Q_PROPERTY(float epsilon READ get_epsilon WRITE set_epsilon RESET reset_epsilon STORED true)
BR_PROPERTY(int, winSize, 80)
BR_PROPERTY(float, C, 1)
BR_PROPERTY(float, epsilon, .01)
private:
typedef scan_fhog_pyramid<pyramid_down<6> > image_scanner_type;
mutable object_detector<image_scanner_type> detector;
mutable QMutex mutex;
void train(const TemplateList &data)
{
dlib::array<array2d<unsigned char> > samples;
std::vector<std::vector<rectangle> > boxes;
foreach (const Template &t, data) {
if (!t.file.rects().isEmpty()) {
cv_image<unsigned char> cimg(t.m());
array2d<unsigned char> image;
assign_image(image,cimg);
samples.push_back(image);
std::vector<rectangle> b;
foreach (const QRectF &r, t.file.rects())
b.push_back(rectangle(r.left(),r.top(),r.right(),r.bottom()));
boxes.push_back(b);
}
}
if (samples.size() == 0)
qFatal("Training data has no bounding boxes.");
image_scanner_type scanner;
scanner.set_detection_window_size(winSize, winSize);
structural_object_detection_trainer<image_scanner_type> trainer(scanner);
trainer.set_num_threads(max(1,QThread::idealThreadCount()));
trainer.set_c(C);
trainer.set_epsilon(epsilon);
if (Globals->verbose)
trainer.be_verbose();
detector = trainer.train(samples, boxes);
}
void project(const Template &src, Template &dst) const
{
dst = src;
cv_image<unsigned char> cimg(src.m());
array2d<unsigned char> image;
assign_image(image,cimg);
QMutexLocker locker(&mutex);
std::vector<rectangle> dets = detector(image);
locker.unlock();
for (size_t i=0; i<dets.size(); i++)
dst.file.appendRect(QRectF(QPointF(dets[i].left(),dets[i].top()),QPointF(dets[i].right(),dets[i].bottom())));
}
void store(QDataStream &stream) const
{
// Create local file
QTemporaryFile tempFile;
tempFile.open();
tempFile.close();
dlib::serialize(qPrintable(tempFile.fileName())) << detector;
// Copy local file contents to stream
tempFile.open();
QByteArray data = tempFile.readAll();
tempFile.close();
stream << data;
}
void load(QDataStream &stream)
{
// Copy local file contents from stream
QByteArray data;
stream >> data;
// Create local file
QTemporaryFile tempFile(QDir::tempPath()+"/model");
tempFile.open();
tempFile.write(data);
tempFile.close();
// Load MLP from local file
dlib::deserialize(qPrintable(tempFile.fileName())) >> detector;
}
};
BR_REGISTER(Transform, DObjectDetectorTransform)
} // namespace br
#include "dlib.moc"