1 #include "perf_precomp.hpp" 2 3 using namespace std; 4 using namespace cv; 5 using namespace perf; 6 using std::tr1::make_tuple; 7 using std::tr1::get; 8 9 CV_ENUM(NormType, NORM_L1, NORM_L2, NORM_L2SQR, NORM_HAMMING, NORM_HAMMING2) 10 11 typedef std::tr1::tuple<NormType, MatType, bool> Norm_Destination_CrossCheck_t; 12 typedef perf::TestBaseWithParam<Norm_Destination_CrossCheck_t> Norm_Destination_CrossCheck; 13 14 typedef std::tr1::tuple<NormType, bool> Norm_CrossCheck_t; 15 typedef perf::TestBaseWithParam<Norm_CrossCheck_t> Norm_CrossCheck; 16 17 typedef std::tr1::tuple<MatType, bool> Source_CrossCheck_t; 18 typedef perf::TestBaseWithParam<Source_CrossCheck_t> Source_CrossCheck; 19 20 void generateData( Mat& query, Mat& train, const int sourceType ); 21 22 PERF_TEST_P(Norm_Destination_CrossCheck, batchDistance_8U, 23 testing::Combine(testing::Values((int)NORM_L1, (int)NORM_L2SQR), 24 testing::Values(CV_32S, CV_32F), 25 testing::Bool() 26 ) 27 ) 28 { 29 NormType normType = get<0>(GetParam()); 30 int destinationType = get<1>(GetParam()); 31 bool isCrossCheck = get<2>(GetParam()); 32 int knn = isCrossCheck ? 1 : 0; 33 34 Mat queryDescriptors; 35 Mat trainDescriptors; 36 Mat dist; 37 Mat ndix; 38 39 generateData(queryDescriptors, trainDescriptors, CV_8U); 40 41 TEST_CYCLE() 42 { 43 batchDistance(queryDescriptors, trainDescriptors, dist, destinationType, (isCrossCheck) ? ndix : noArray(), 44 normType, knn, Mat(), 0, isCrossCheck); 45 } 46 47 SANITY_CHECK(dist); 48 if (isCrossCheck) SANITY_CHECK(ndix); 49 } 50 51 PERF_TEST_P(Norm_CrossCheck, batchDistance_Dest_32S, 52 testing::Combine(testing::Values((int)NORM_HAMMING, (int)NORM_HAMMING2), 53 testing::Bool() 54 ) 55 ) 56 { 57 NormType normType = get<0>(GetParam()); 58 bool isCrossCheck = get<1>(GetParam()); 59 int knn = isCrossCheck ? 1 : 0; 60 61 Mat queryDescriptors; 62 Mat trainDescriptors; 63 Mat dist; 64 Mat ndix; 65 66 generateData(queryDescriptors, trainDescriptors, CV_8U); 67 68 TEST_CYCLE() 69 { 70 batchDistance(queryDescriptors, trainDescriptors, dist, CV_32S, (isCrossCheck) ? ndix : noArray(), 71 normType, knn, Mat(), 0, isCrossCheck); 72 } 73 74 SANITY_CHECK(dist); 75 if (isCrossCheck) SANITY_CHECK(ndix); 76 } 77 78 PERF_TEST_P(Source_CrossCheck, batchDistance_L2, 79 testing::Combine(testing::Values(CV_8U, CV_32F), 80 testing::Bool() 81 ) 82 ) 83 { 84 int sourceType = get<0>(GetParam()); 85 bool isCrossCheck = get<1>(GetParam()); 86 int knn = isCrossCheck ? 1 : 0; 87 88 Mat queryDescriptors; 89 Mat trainDescriptors; 90 Mat dist; 91 Mat ndix; 92 93 generateData(queryDescriptors, trainDescriptors, sourceType); 94 95 declare.time(50); 96 TEST_CYCLE() 97 { 98 batchDistance(queryDescriptors, trainDescriptors, dist, CV_32F, (isCrossCheck) ? ndix : noArray(), 99 NORM_L2, knn, Mat(), 0, isCrossCheck); 100 } 101 102 SANITY_CHECK(dist); 103 if (isCrossCheck) SANITY_CHECK(ndix); 104 } 105 106 PERF_TEST_P(Norm_CrossCheck, batchDistance_32F, 107 testing::Combine(testing::Values((int)NORM_L1, (int)NORM_L2SQR), 108 testing::Bool() 109 ) 110 ) 111 { 112 NormType normType = get<0>(GetParam()); 113 bool isCrossCheck = get<1>(GetParam()); 114 int knn = isCrossCheck ? 1 : 0; 115 116 Mat queryDescriptors; 117 Mat trainDescriptors; 118 Mat dist; 119 Mat ndix; 120 121 generateData(queryDescriptors, trainDescriptors, CV_32F); 122 declare.time(100); 123 124 TEST_CYCLE() 125 { 126 batchDistance(queryDescriptors, trainDescriptors, dist, CV_32F, (isCrossCheck) ? ndix : noArray(), 127 normType, knn, Mat(), 0, isCrossCheck); 128 } 129 130 SANITY_CHECK(dist, 1e-4); 131 if (isCrossCheck) SANITY_CHECK(ndix); 132 } 133 134 void generateData( Mat& query, Mat& train, const int sourceType ) 135 { 136 const int dim = 500; 137 const int queryDescCount = 300; // must be even number because we split train data in some cases in two 138 const int countFactor = 4; // do not change it 139 RNG& rng = theRNG(); 140 141 // Generate query descriptors randomly. 142 // Descriptor vector elements are integer values. 143 Mat buf( queryDescCount, dim, CV_32SC1 ); 144 rng.fill( buf, RNG::UNIFORM, Scalar::all(0), Scalar(3) ); 145 buf.convertTo( query, sourceType ); 146 147 // Generate train decriptors as follows: 148 // copy each query descriptor to train set countFactor times 149 // and perturb some one element of the copied descriptors in 150 // in ascending order. General boundaries of the perturbation 151 // are (0.f, 1.f). 152 train.create( query.rows*countFactor, query.cols, sourceType ); 153 float step = (sourceType == CV_8U ? 256.f : 1.f) / countFactor; 154 for( int qIdx = 0; qIdx < query.rows; qIdx++ ) 155 { 156 Mat queryDescriptor = query.row(qIdx); 157 for( int c = 0; c < countFactor; c++ ) 158 { 159 int tIdx = qIdx * countFactor + c; 160 Mat trainDescriptor = train.row(tIdx); 161 queryDescriptor.copyTo( trainDescriptor ); 162 int elem = rng(dim); 163 float diff = rng.uniform( step*c, step*(c+1) ); 164 trainDescriptor.col(elem) += diff; 165 } 166 } 167 } 168
