从零学做网站,西安网站价格,黄村做网站建设,市场营销策划方案3000字先把模型放在如下目录 运行如下代码
import cv2
import numpy as npclass Onnx_clf:def __init__(self, onnx:strdnn_model1/plane02.onnx, img_size640, classlist:list[plane]) - None: func: 读取onnx模型,并进行目标识别para onnx:模型路径img_size:输出图片大小,和模…先把模型放在如下目录 运行如下代码
import cv2
import numpy as npclass Onnx_clf:def __init__(self, onnx:strdnn_model1/plane02.onnx, img_size640, classlist:list[plane]) - None: func: 读取onnx模型,并进行目标识别para onnx:模型路径img_size:输出图片大小,和模型直接相关classlist:类别列表return: Noneself.net cv2.dnn.readNet(onnx) # 读取模型self.img_size img_size # 输出图片尺寸大小self.classlist classlist # 读取类别列表def img_identify(self, img, ifshowTrue) - np.ndarray: func: 图片识别para img: 图片路径或者图片数组ifshow: 是否显示图片return: 图片数组if type(img) str: src cv2.imread(img)else: src imgheight, width, _ src.shape #注意输出的尺寸是先高后宽_max max(width, height)resized np.zeros((_max, _max, 3), np.uint8)resized[0:height, 0:width] src # 将图片转换成正方形防止后续图片预处理(缩放)失真# 图像预处理函数,缩放裁剪,交换通道 img scale out_size swapRBblob cv2.dnn.blobFromImage(resized, 1/255.0, (self.img_size, self.img_size), swapRBTrue)prop _max / self.img_size # 计算缩放比例dst cv2.resize(src, (round(width/prop), round(height/prop)))# print(prop) # 注意这里不能取整而是需要取小数否则后面绘制框的时候会出现偏差self.net.setInput(blob) # 将图片输入到模型out self.net.forward() # 模型输出# print(out.shape)out np.array(out[0])out out[out[:, 4] 0.5] # 利用numpy的花式索引,速度更快, 过滤置信度低的目标boxes out[:, :4]confidences out[:, 4]class_ids np.argmax(out[:, 5:], axis1)class_scores np.max(out[:, 5:], axis1)# out2 out[0][out[0][:][4] 0.5]# for i in out[0]: # 遍历每一个框# class_max_score max(i[5:])# if i[4] 0.5 or class_max_score 0.25: # 过滤置信度低的目标# continue# boxes.append(i[:4]) # 获取目标框: x,y,w,h (x,y为中心点坐标)# confidences.append(i[4]) # 获取置信度# class_ids.append(np.argmax(i[5:])) # 获取类别id# class_scores.append(class_max_score) # 获取类别置信度indexes cv2.dnn.NMSBoxes(boxes, confidences, 0.25, 0.45) # 非极大值抑制, 获取的是索引# print(indexes)iffall True if len(indexes)!0 else False# print(iffall)for i in indexes: # 遍历每一个目标, 绘制目标框box boxes[i]class_id class_ids[i]score round(class_scores[i], 2)x1 round((box[0] - 0.5*box[2])*prop)y1 round((box[1] - 0.5*box[3])*prop)x2 round((box[0] 0.5*box[2])*prop)y2 round((box[1] 0.5*box[3])*prop)# print(x1, y1, x2, y2)self.drawtext(src,(x1, y1), (x2, y2), self.classlist[class_id] str(score))dst cv2.resize(src, (round(width/prop), round(height/prop)))if ifshow:cv2.imshow(result, dst)cv2.waitKey(0)return dst, iffalldef video_identify(self, video_path:str) - None: func: 视频识别para video_path: 视频路径return: Nonecap cv2.VideoCapture(video_path)fps cap.get(cv2.CAP_PROP_FPS)# print(fps)while cap.isOpened():ret, frame cap.read()#键盘输入空格暂停输入q退出key cv2.waitKey(1) 0xffif key ord( ): cv2.waitKey(0)if key ord(q): breakif not ret: breakimg, res self.img_identify(frame, False)cv2.imshow(result, img)print(res)if cv2.waitKey(int(1000/fps)) ord(q):breakcap.release()cv2.destroyAllWindows()staticmethoddef drawtext(image, pt1, pt2, text): func: 根据给出的坐标和文本,在图片上进行绘制para image: 图片数组; pt1: 左上角坐标; pt2: 右下角坐标; text: 矩形框上显示的文本,即类别信息return: NonefontFace cv2.FONT_HERSHEY_COMPLEX_SMALL # 字体# fontFace cv2.FONT_HERSHEY_COMPLEX # 字体fontScale 1.5 # 字体大小line_thickness 3 # 线条粗细font_thickness 2 # 文字笔画粗细line_back_color (0, 0, 255) # 线条和文字背景颜色:红色font_color (255, 255, 255) # 文字颜色:白色# 绘制矩形框cv2.rectangle(image, pt1, pt2, colorline_back_color, thicknessline_thickness)# 计算文本的宽高: retval:文本的宽高; baseLine:基线与最低点之间的距离(本例未使用)retval, baseLine cv2.getTextSize(text,fontFacefontFace,fontScalefontScale, thicknessfont_thickness)# 计算覆盖文本的矩形框坐标topleft (pt1[0], pt1[1] - retval[1]) # 基线与目标框上边缘重合(不考虑基线以下的部分)bottomright (topleft[0] retval[0], topleft[1] retval[1])cv2.rectangle(image, topleft, bottomright, thickness-1, colorline_back_color) # 绘制矩形框(填充)# 绘制文本cv2.putText(image, text, pt1, fontScalefontScale,fontFacefontFace, colorfont_color, thicknessfont_thickness)if __name__ __main__:clf Onnx_clf()import tkinter as tkfrom tkinter.filedialog import askopenfilenametk.Tk().withdraw() # 隐藏主窗口, 必须要用否则会有一个小窗口source askopenfilename(titletest2.mp4)# source rC:\Users\Zeoy\Desktop\YOLOData\data\IMG_568.jpgif source.endswith(.jpg) or source.endswith(.png) or source.endswith(.bmp):res, out clf.img_identify(source, False)print(out)cv2.imshow(result, res)cv2.waitKey(0)elif source.endswith(.mp4) or source.endswith(.avi):print(视频识别中...按q退出)clf.video_identify(source)else:print(不支持的文件格式)
