神经网络不会举一反三，郁闷。。。
1
////////////////////////////////////////////////////////////////////////
//
2 // File Name: pnn.cpp
3 // Author: Ruoruo(du#in.tum.de)
4 //////////////////////////////////////////////////////////////////////// //
5 #include " stdafx.h "
6 #include " cv.h "
7 #include " highgui.h "
8 #include < ml.h >
9 #include < time.h >
10 #include < ctype.h >
11 #include < vector >
12 #include < math.h >
13 #include < iostream >
14 using namespace std;
15
16 static CvScalar colors[] =
17 {
18 {{ 0 , 0 , 255 }},
19 {{ 0 , 128 , 255 }},
20 {{ 0 , 255 , 255 }},
21 {{ 0 , 255 , 0 }},
22 {{ 255 , 128 , 0 }},
23 {{ 255 , 255 , 0 }},
24 {{ 255 , 0 , 0 }},
25 {{ 255 , 0 , 255 }}
26 };
27
28 int main( int argc, char ** argv )
29 {
30 vector < float > point;
31 vector < float > result;
32
33 float p[ 10 ] = { 1.3 , 2.7 ,
34 1.5 , 3.0 ,
35 1.7 , 2.8 ,
36 1.6 , 2.6 ,
37 1.2 , 2.9 };
38 float res[ 5 ] = { 0 , 0 , 0 , 0 , 0 };
39 int i;
40 for (i = 0 ; i < 10 ; i ++ )
41 {
42 point.push_back(p[i]);
43 if (i < 5 ) result.push_back(res[i]);
44 }
45
46 CvMat * input = cvCreateMat( 5 , 2 , CV_32FC1 );
47 cvInitMatHeader( input, 5 , 2 , CV_32FC1, p );
48 CvMat * output = cvCreateMat( 5 , 1 , CV_32FC1 );
49 cvInitMatHeader( output, 5 , 1 , CV_32FC1, res );
50 IplImage * img = cvCreateImage(cvSize( 450 , 450 ), IPL_DEPTH_8U, 3 );
51 img -> origin = 1 ;
52 for (i = 0 ; i < 5 ; i ++ )
53 {
54 cvCircle(img, cvPoint(( int )(p[i * 2 ] * 100 ), ( int )(p[i * 2 + 1 ] * 100 )), 5 , colors[( int )res[i] % 8 ], 1 , CV_AA, 0 );
55 }
56
57 int layer_num[ 3 ] = { 2 , 4 , 1 };
58 CvMat * layer_size = cvCreateMatHeader( 1 , 3 , CV_32S );
59 cvInitMatHeader( layer_size, 1 , 3 , CV_32S, layer_num );
60 CvANN_MLP pnn;
61 pnn.create( layer_size, CvANN_MLP::SIGMOID_SYM, 1 , 1 );
62 CvANN_MLP_TrainParams params ;
63 params .term_crit = cvTermCriteria( CV_TERMCRIT_ITER | CV_TERMCRIT_EPS, \
64 300 , 0.0000001 );
65 params .train_method = 0 ;
66 params .bp_dw_scale = 0.1 ;
67 params .bp_moment_scale = 0.1 ;
68 cout << " begin training
"
<<
endl;
69 pnn.train( input, output, 0 , 0 , params );
70 cout << " end training
"
<<
endl;
71 pnn.save( " pNN_DATA.xml " );
72
73 // begin to test
74 float testp[ 24 ] = { 1.4 , 2.75 ,
75 4.3 , 0.2 ,
76 4.2 , 4.3 ,
77 1.45 , 2.85 ,
78 4.2 , 0.4 ,
79 4.1 , 4.0 ,
80 4.3 , 0.5 ,
81 4.0 , 4.2 ,
82 1.5 , 2.7 ,
83 4.1 , 4.2 ,
84 1.6 , 2.7 ,
85 4.0 , 0.3 };
86 /* float testp[24] = { 1.4, 2.75,
87 4.3, 0.2,
88 4.2, 4.3,
89 1.45, 2.85,
90 4.25, 0.3,
91 4.25, 4.25,
92 3.5, 1.2,
93 3.0, 3.7,
94 4.0, 2.7,
95 0.2, 0.2,
96 2.8, 2.7,
97 2.7, 2.8 }; */
98 CvMat * test_point = cvCreateMat( 1 , 2 , CV_32FC1 );
99 CvMat * test_result = cvCreateMat( 1 , 1 , CV_32FC1 );
100 CvFont font;
101 double hScale = 0.5 ;
102 double vScale = 0.5 ;
103 int lineWidth = 1 ;
104 cvInitFont( & font, CV_FONT_HERSHEY_COMPLEX | CV_FONT_ITALIC, hScale,vScale, 0 ,lineWidth);
105
106 for (i = 0 ; i < 12 ; i ++ )
107 {
108 cvSetReal2D( test_point, 0 , 0 , testp[ 2 * i] );
109 cvSetReal2D( test_point, 0 , 1 , testp[ 2 * i + 1 ] );
110 pnn.predict(test_point, test_result);
111 cout << cvmGet(test_result, 0 , 0 ) << endl;
112
113 float delta = 1 ;
114 int best_class = 0 ;
115 int max_class = 0 ;
116 for ( int ii = 0 ; ii < result.size(); ii ++ ){
117 if (fabs(cvmGet(test_result, 0 , 0 ) - ( float )result[ii]) < delta){
118 delta = fabs(cvmGet(test_result, 0 , 0 ) - ( float )result[ii]);
119 best_class = result[ii];
120 }
121 if (result[ii] >= max_class)
122 max_class = result[ii];
123 }
124
125 point.push_back(testp[ 2 * i]);
126 point.push_back(testp[ 2 * i + 1 ]);
127
128 if ( delta > 0.06 ){
129 int new_result = max_class + 1 ;
130 cvmSet( test_result, 0 , 0 ,new_result );
131 result.push_back(( float )new_result );
132 }
133 else {
134 cvmSet( test_result, 0 , 0 ,best_class );
135 result.push_back(( float )best_class );
136 }
137
138 int new_point_size = point.size();
139 int new_result_size = result.size();
140
141 CvMat * input = cvCreateMat( new_result_size, 2 , CV_32FC1 );
142 CvMat * output = cvCreateMat( new_result_size, 1 , CV_32FC1 );
143
144 for ( int ii = 0 ; ii < new_result_size; ii ++ )
145 {
146 cvmSet( input, ii, 0 , point[ 2 * ii]);
147 cvmSet( input, ii, 1 , point[ 2 * ii + 1 ]);
148 cvmSet( output, ii, 0 , result[ii]);
149 }
150 // cout<<"begin training again"<<endl;
151 pnn.train( input, output, 0 , 0 , params );
152 // cout<<"end training"<<endl;
153
154 cvCircle( img, cvPoint(( int )(testp[i * 2 ] * 100 ), ( int )(testp[i * 2 + 1 ] * 100 )), 0 , colors[( int )cvmGet(test_result, 0 , 0 ) % 8 ], 10 , CV_AA, 0 );
155
156 char buffer[ 10 ];
157 _itoa(i,buffer, 10 );
158 string point_id(buffer);
159 cvPutText(img, point_id.c_str(), cvPoint(testp[ 2 * i] * 100 ,testp[ 2 * i + 1 ] * 100 ), & font, cvScalar( 255 , 255 , 255 ));
160
161 cout << i << " : " << " ( " << testp[i * 2 ] << " , " << testp[i * 2 + 1 ] << " ) " << " \t " << cvmGet(test_result, 0 , 0 ) << endl;
162 }
163
164 cvNamedWindow( " Coordinates " , 1 );
165 cvShowImage( " Coordinates " ,img);
166
167 cvWaitKey( 0 );
168
169 cvDestroyWindow( " Coordinates " );
170 cvReleaseImage( & img);
171
172 return 0 ;
173 }
2 // File Name: pnn.cpp
3 // Author: Ruoruo(du#in.tum.de)
4 //////////////////////////////////////////////////////////////////////// //
5 #include " stdafx.h "
6 #include " cv.h "
7 #include " highgui.h "
8 #include < ml.h >
9 #include < time.h >
10 #include < ctype.h >
11 #include < vector >
12 #include < math.h >
13 #include < iostream >
14 using namespace std;
15
16 static CvScalar colors[] =
17 {
18 {{ 0 , 0 , 255 }},
19 {{ 0 , 128 , 255 }},
20 {{ 0 , 255 , 255 }},
21 {{ 0 , 255 , 0 }},
22 {{ 255 , 128 , 0 }},
23 {{ 255 , 255 , 0 }},
24 {{ 255 , 0 , 0 }},
25 {{ 255 , 0 , 255 }}
26 };
27
28 int main( int argc, char ** argv )
29 {
30 vector < float > point;
31 vector < float > result;
32
33 float p[ 10 ] = { 1.3 , 2.7 ,
34 1.5 , 3.0 ,
35 1.7 , 2.8 ,
36 1.6 , 2.6 ,
37 1.2 , 2.9 };
38 float res[ 5 ] = { 0 , 0 , 0 , 0 , 0 };
39 int i;
40 for (i = 0 ; i < 10 ; i ++ )
41 {
42 point.push_back(p[i]);
43 if (i < 5 ) result.push_back(res[i]);
44 }
45
46 CvMat * input = cvCreateMat( 5 , 2 , CV_32FC1 );
47 cvInitMatHeader( input, 5 , 2 , CV_32FC1, p );
48 CvMat * output = cvCreateMat( 5 , 1 , CV_32FC1 );
49 cvInitMatHeader( output, 5 , 1 , CV_32FC1, res );
50 IplImage * img = cvCreateImage(cvSize( 450 , 450 ), IPL_DEPTH_8U, 3 );
51 img -> origin = 1 ;
52 for (i = 0 ; i < 5 ; i ++ )
53 {
54 cvCircle(img, cvPoint(( int )(p[i * 2 ] * 100 ), ( int )(p[i * 2 + 1 ] * 100 )), 5 , colors[( int )res[i] % 8 ], 1 , CV_AA, 0 );
55 }
56
57 int layer_num[ 3 ] = { 2 , 4 , 1 };
58 CvMat * layer_size = cvCreateMatHeader( 1 , 3 , CV_32S );
59 cvInitMatHeader( layer_size, 1 , 3 , CV_32S, layer_num );
60 CvANN_MLP pnn;
61 pnn.create( layer_size, CvANN_MLP::SIGMOID_SYM, 1 , 1 );
62 CvANN_MLP_TrainParams params ;
63 params .term_crit = cvTermCriteria( CV_TERMCRIT_ITER | CV_TERMCRIT_EPS, \
64 300 , 0.0000001 );
65 params .train_method = 0 ;
66 params .bp_dw_scale = 0.1 ;
67 params .bp_moment_scale = 0.1 ;
68 cout << " begin training
"
<<
endl;69 pnn.train( input, output, 0 , 0 , params );
70 cout << " end training
"
<<
endl;71 pnn.save( " pNN_DATA.xml " );
72
73 // begin to test
74 float testp[ 24 ] = { 1.4 , 2.75 ,
75 4.3 , 0.2 ,
76 4.2 , 4.3 ,
77 1.45 , 2.85 ,
78 4.2 , 0.4 ,
79 4.1 , 4.0 ,
80 4.3 , 0.5 ,
81 4.0 , 4.2 ,
82 1.5 , 2.7 ,
83 4.1 , 4.2 ,
84 1.6 , 2.7 ,
85 4.0 , 0.3 };
86 /* float testp[24] = { 1.4, 2.75,
87 4.3, 0.2,
88 4.2, 4.3,
89 1.45, 2.85,
90 4.25, 0.3,
91 4.25, 4.25,
92 3.5, 1.2,
93 3.0, 3.7,
94 4.0, 2.7,
95 0.2, 0.2,
96 2.8, 2.7,
97 2.7, 2.8 }; */
98 CvMat * test_point = cvCreateMat( 1 , 2 , CV_32FC1 );
99 CvMat * test_result = cvCreateMat( 1 , 1 , CV_32FC1 );
100 CvFont font;
101 double hScale = 0.5 ;
102 double vScale = 0.5 ;
103 int lineWidth = 1 ;
104 cvInitFont( & font, CV_FONT_HERSHEY_COMPLEX | CV_FONT_ITALIC, hScale,vScale, 0 ,lineWidth);
105
106 for (i = 0 ; i < 12 ; i ++ )
107 {
108 cvSetReal2D( test_point, 0 , 0 , testp[ 2 * i] );
109 cvSetReal2D( test_point, 0 , 1 , testp[ 2 * i + 1 ] );
110 pnn.predict(test_point, test_result);
111 cout << cvmGet(test_result, 0 , 0 ) << endl;
112
113 float delta = 1 ;
114 int best_class = 0 ;
115 int max_class = 0 ;
116 for ( int ii = 0 ; ii < result.size(); ii ++ ){
117 if (fabs(cvmGet(test_result, 0 , 0 ) - ( float )result[ii]) < delta){
118 delta = fabs(cvmGet(test_result, 0 , 0 ) - ( float )result[ii]);
119 best_class = result[ii];
120 }
121 if (result[ii] >= max_class)
122 max_class = result[ii];
123 }
124
125 point.push_back(testp[ 2 * i]);
126 point.push_back(testp[ 2 * i + 1 ]);
127
128 if ( delta > 0.06 ){
129 int new_result = max_class + 1 ;
130 cvmSet( test_result, 0 , 0 ,new_result );
131 result.push_back(( float )new_result );
132 }
133 else {
134 cvmSet( test_result, 0 , 0 ,best_class );
135 result.push_back(( float )best_class );
136 }
137
138 int new_point_size = point.size();
139 int new_result_size = result.size();
140
141 CvMat * input = cvCreateMat( new_result_size, 2 , CV_32FC1 );
142 CvMat * output = cvCreateMat( new_result_size, 1 , CV_32FC1 );
143
144 for ( int ii = 0 ; ii < new_result_size; ii ++ )
145 {
146 cvmSet( input, ii, 0 , point[ 2 * ii]);
147 cvmSet( input, ii, 1 , point[ 2 * ii + 1 ]);
148 cvmSet( output, ii, 0 , result[ii]);
149 }
150 // cout<<"begin training again
"<<endl;
151 pnn.train( input, output, 0 , 0 , params );
152 // cout<<"end training
"<<endl;
153
154 cvCircle( img, cvPoint(( int )(testp[i * 2 ] * 100 ), ( int )(testp[i * 2 + 1 ] * 100 )), 0 , colors[( int )cvmGet(test_result, 0 , 0 ) % 8 ], 10 , CV_AA, 0 );
155
156 char buffer[ 10 ];
157 _itoa(i,buffer, 10 );
158 string point_id(buffer);
159 cvPutText(img, point_id.c_str(), cvPoint(testp[ 2 * i] * 100 ,testp[ 2 * i + 1 ] * 100 ), & font, cvScalar( 255 , 255 , 255 ));
160
161 cout << i << " : " << " ( " << testp[i * 2 ] << " , " << testp[i * 2 + 1 ] << " ) " << " \t " << cvmGet(test_result, 0 , 0 ) << endl;
162 }
163
164 cvNamedWindow( " Coordinates " , 1 );
165 cvShowImage( " Coordinates " ,img);
166
167 cvWaitKey( 0 );
168
169 cvDestroyWindow( " Coordinates " );
170 cvReleaseImage( & img);
171
172 return 0 ;
173 }