从opencv3.4.9中摘取。
写svm的xml信息：
1 void write( FileStorage& fs ) const CV_OVERRIDE 2 { 3 int class_count = !class_labels.empty() ? (int)class_labels.total() : 4 params.svmType == ONE_CLASS ? 1 : 0; 5 if( !isTrained() ) 6 CV_Error( CV_StsParseError, "SVM model data is invalid, check sv_count, var_* and class_count tags" ); 7 8 writeFormat(fs); 9 write_params( fs ); 10 11 fs << "var_count" << var_count; 12 13 if( class_count > 0 ) 14 { 15 fs << "class_count" << class_count; 16 17 if( !class_labels.empty() ) 18 fs << "class_labels" << class_labels; 19 20 if( !params.classWeights.empty() ) 21 fs << "class_weights" << params.classWeights; 22 } 23 24 // write the joint collection of support vectors 25 int i, sv_total = sv.rows; 26 fs << "sv_total" << sv_total; 27 fs << "support_vectors" << "["; 28 for( i = 0; i < sv_total; i++ ) 29 { 30 fs << "[:"; 31 fs.writeRaw("f", sv.ptr(i), sv.cols*sv.elemSize()); 32 fs << "]"; 33 } 34 fs << "]"; 35 36 if ( !uncompressed_sv.empty() ) 37 { 38 // write the joint collection of uncompressed support vectors 39 int uncompressed_sv_total = uncompressed_sv.rows; 40 fs << "uncompressed_sv_total" << uncompressed_sv_total; 41 fs << "uncompressed_support_vectors" << "["; 42 for( i = 0; i < uncompressed_sv_total; i++ ) 43 { 44 fs << "[:"; 45 fs.writeRaw("f", uncompressed_sv.ptr(i), uncompressed_sv.cols*uncompressed_sv.elemSize()); 46 fs << "]"; 47 } 48 fs << "]"; 49 } 50 51 // write decision functions 52 int df_count = (int)decision_func.size(); 53 54 fs << "decision_functions" << "["; 55 for( i = 0; i < df_count; i++ ) 56 { 57 const DecisionFunc& df = decision_func[i]; 58 int sv_count = getSVCount(i); 59 fs << "{" << "sv_count" << sv_count 60 << "rho" << df.rho 61 << "alpha" << "[:"; 62 fs.writeRaw("d", (const uchar*)&df_alpha[df.ofs], sv_count*sizeof(df_alpha[0])); 63 fs << "]"; 64 if( class_count >= 2 ) 65 { 66 fs << "index" << "[:"; 67 fs.writeRaw("i", (const uchar*)&df_index[df.ofs], sv_count*sizeof(df_index[0])); 68 fs << "]"; 69 } 70 else 71 CV_Assert( sv_count == sv_total ); 72 fs << "}"; 73 } 74 fs << "]"; 75 }
读svm的xml信息：
1 void read( const FileNode& fn ) CV_OVERRIDE 2 { 3 clear(); 4 5 // read SVM parameters 6 read_params( fn ); 7 8 // and top-level data 9 int i, sv_total = (int)fn["sv_total"]; 10 var_count = (int)fn["var_count"]; 11 int class_count = (int)fn["class_count"]; 12 13 if( sv_total <= 0 || var_count <= 0 ) 14 CV_Error( CV_StsParseError, "SVM model data is invalid, check sv_count, var_* and class_count tags" ); 15 16 FileNode m = fn["class_labels"]; 17 if( !m.empty() ) 18 m >> class_labels; 19 m = fn["class_weights"]; 20 if( !m.empty() ) 21 m >> params.classWeights; 22 23 if( class_count > 1 && (class_labels.empty() || (int)class_labels.total() != class_count)) 24 CV_Error( CV_StsParseError, "Array of class labels is missing or invalid" ); 25 26 // read support vectors 27 FileNode sv_node = fn["support_vectors"]; 28 29 CV_Assert((int)sv_node.size() == sv_total); 30 31 sv.create(sv_total, var_count, CV_32F); 32 FileNodeIterator sv_it = sv_node.begin(); 33 for( i = 0; i < sv_total; i++, ++sv_it ) 34 { 35 (*sv_it).readRaw("f", sv.ptr(i), var_count*sv.elemSize()); 36 } 37 38 int uncompressed_sv_total = (int)fn["uncompressed_sv_total"]; 39 40 if( uncompressed_sv_total > 0 ) 41 { 42 // read uncompressed support vectors 43 FileNode uncompressed_sv_node = fn["uncompressed_support_vectors"]; 44 45 CV_Assert((int)uncompressed_sv_node.size() == uncompressed_sv_total); 46 uncompressed_sv.create(uncompressed_sv_total, var_count, CV_32F); 47 48 FileNodeIterator uncompressed_sv_it = uncompressed_sv_node.begin(); 49 for( i = 0; i < uncompressed_sv_total; i++, ++uncompressed_sv_it ) 50 { 51 (*uncompressed_sv_it).readRaw("f", uncompressed_sv.ptr(i), var_count*uncompressed_sv.elemSize()); 52 } 53 } 54 55 // read decision functions 56 int df_count = class_count > 1 ? class_count*(class_count-1)/2 : 1; 57 FileNode df_node = fn["decision_functions"]; 58 59 CV_Assert((int)df_node.size() == df_count); 60 61 FileNodeIterator df_it = df_node.begin(); 62 for( i = 0; i < df_count; i++, ++df_it ) 63 { 64 FileNode dfi = *df_it; 65 DecisionFunc df; 66 int sv_count = (int)dfi["sv_count"]; 67 int ofs = (int)df_index.size(); 68 df.rho = (double)dfi["rho"]; 69 df.ofs = ofs; 70 df_index.resize(ofs + sv_count); 71 df_alpha.resize(ofs + sv_count); 72 dfi["alpha"].readRaw("d", (uchar*)&df_alpha[ofs], sv_count*sizeof(df_alpha[0])); 73 if( class_count >= 2 ) 74 dfi["index"].readRaw("i", (uchar*)&df_index[ofs], sv_count*sizeof(df_index[0])); 75 decision_func.push_back(df); 76 } 77 if( class_count < 2 ) 78 setRangeVector(df_index, sv_total); 79 if( (int)fn["optimize_linear"] != 0 ) 80 optimize_linear_svm(); 81 } 82 83 SvmParams params; 84 Mat class_labels; 85 int var_count; 86 Mat sv, uncompressed_sv; 87 vector<DecisionFunc> decision_func; 88 vector<double> df_alpha; 89 vector<int> df_index; 90 91 Ptr<Kernel> kernel; 92 };