get class-length from output-size instead of class_names.length

Author: UNeedCryDear

yolov5.cpp

@@ -52,15 +52,16 @@ bool Yolov5::Detect(Mat& SrcImg, Net& net, vector<OutputSeg>& output) {
 	std::vector<cv::Rect> boxes;//每个id矩形框
 	float ratio_h = (float)netInputImg.rows / _netHeight;
 	float ratio_w = (float)netInputImg.cols / _netWidth;
-	int net_width = _className.size() + 5;  //输出的网络宽度是类别数+5
-	int net_out_width = netOutputImg[0].size[2];
-	assert(net_out_width == net_width, "Error Wrong number of _className");  //模型类别数目不对
+	int net_width = netOutputImg[0].size[2];  //输出的网络宽度是类别数+5
+	//int net_out_width = netOutputImg[0].size[2];
+	//assert(net_out_width == net_width, "Error Wrong number of _className");  //模型类别数目不对
 	float* pdata = (float*)netOutputImg[0].data;
 	int net_height = netOutputImg[0].size[1];
+	int score_length = net_width - 5;
 	for (int r = 0; r < net_height; ++r) {  
 		float box_score = pdata[4]; ;//获取每一行的box框中含有某个物体的概率
 		if (box_score >= _classThreshold) {
-			cv::Mat scores(1, _className.size(), CV_32FC1, pdata + 5);
+			cv::Mat scores(1, score_length, CV_32FC1, pdata + 5);
 			Point classIdPoint;
 			double max_class_socre;
 			minMaxLoc(scores, 0, &max_class_socre, 0, &classIdPoint);

yolov5_onnx.cpp

@@ -198,14 +198,15 @@ bool Yolov5Onnx::OnnxBatchDetect(std::vector<cv::Mat>& srcImgs, std::vector<std:
 	int64_t one_output_length = VectorProduct(_outputTensorShape) / _outputTensorShape[0];
 	int net_width = _outputTensorShape[2];
 	int net_height = _outputTensorShape[1];
+	int score_length = net_width - 5;
 	for (int img_index = 0; img_index < srcImgs.size(); ++img_index) {
 		std::vector<int> class_ids;//结果id数组
 		std::vector<float> confidences;//结果每个id对应置信度数组
 		std::vector<cv::Rect> boxes;//每个id矩形框
 		for (int r = 0; r < net_height; r++) {    //stride
 			float box_score = pdata[4]; ;//box-confidence
 			if (box_score >= _classThreshold) {
-				cv::Mat scores(1, _className.size(), CV_32FC1, pdata + 5);
+				cv::Mat scores(1, score_length, CV_32FC1, pdata + 5);
 				Point classIdPoint;
 				double max_class_socre;
 				minMaxLoc(scores, 0, &max_class_socre, 0, &classIdPoint);

yolov5_seg.cpp

@@ -65,18 +65,19 @@ bool Yolov5Seg::Detect(Mat& srcImg, Net& net, vector<OutputSeg>& output) {
 	std::vector<vector<float>> picked_proposals;  //output0[:,:, 5 + _className.size():net_width]===> for mask
 	int net_width = net_output_img[0].size[2];
 	int net_height = net_output_img[0].size[1];
+	int score_length = net_width - 37;
 	float* pdata = (float*)net_output_img[0].data;
 	for (int r = 0; r < net_height; r++) {    //lines
 		float box_score = pdata[4];
 		if (box_score >= _classThreshold) {
-			cv::Mat scores(1, _className.size(), CV_32FC1, pdata + 5);
+			cv::Mat scores(1, score_length, CV_32FC1, pdata + 5);
 			Point classIdPoint;
 			double max_class_socre;
 			minMaxLoc(scores, 0, &max_class_socre, 0, &classIdPoint);
 			max_class_socre = (float)max_class_socre;
 			if (max_class_socre >= _classThreshold) {
 
-				vector<float> temp_proto(pdata + 5 + _className.size(), pdata + net_width);
+				vector<float> temp_proto(pdata + 5 + score_length, pdata + net_width);
 				picked_proposals.push_back(temp_proto);
 				//rect [x,y,w,h]
 				float x = (pdata[0] - params[2]) / params[0];  //x

yolov5_seg_onnx.cpp

@@ -225,6 +225,7 @@ bool Yolov5SegOnnx::OnnxBatchDetect(std::vector<cv::Mat>& srcImgs, std::vector<s
 	int64_t one_output_length = VectorProduct(_outputTensorShape) / _outputTensorShape[0];
 	int net_width = _outputTensorShape[2];
 	int net_height = _outputTensorShape[1];
+	int score_length = net_width - 37;
 	for (int img_index = 0; img_index < srcImgs.size(); ++img_index) {
 		std::vector<int> class_ids;//结果id数组
 		std::vector<float> confidences;//结果每个id对应置信度数组
@@ -233,13 +234,13 @@ bool Yolov5SegOnnx::OnnxBatchDetect(std::vector<cv::Mat>& srcImgs, std::vector<s
 		for (int r = 0; r < net_height; r++) {    //stride
 			float box_score = pdata[4]; ;//box-confidence
 			if (box_score >= _classThreshold) {
-				cv::Mat scores(1, _className.size(), CV_32FC1, pdata + 5);
+				cv::Mat scores(1, score_length, CV_32FC1, pdata + 5);
 				Point classIdPoint;
 				double max_class_socre;
 				minMaxLoc(scores, 0, &max_class_socre, 0, &classIdPoint);
 				max_class_socre = (float)max_class_socre;
 				if (max_class_socre >= _classThreshold) {
-					vector<float> temp_proto(pdata + 5 + _className.size(), pdata + net_width);
+					vector<float> temp_proto(pdata + 5 + score_length, pdata + net_width);
 					picked_proposals.push_back(temp_proto);
 					//rect [x,y,w,h]
 					float x = (pdata[0] - params[img_index][2]) / params[img_index][0];  //x