""" 密度估计 """ import cv2 import json import random from configs.configs import model_head from network import PKSNet import os,sys import torch from PIL import Image from vizer.draw import draw_boxes from ssd.config import cfg from ssd.data.datasets import VOCDataset import numpy as np from ssd.data.transforms import build_transforms from ssd.modeling.detector import build_detection_model from ssd.utils.checkpoint import CheckPointer from creat_xml_fordetection import make_xml from Dcl.models.LoadModel import MainModel from Dcl.DclConfig import LoadConfig, load_data_transformers from torch.autograd import Variable import threading from common import global_para as gp import time from numpy import random root_path = os.getcwd() sys.path.insert(0,root_path+"/Y5") from Y5.models.experimental import attempt_load from Y5.utils.datasets import letterbox from Y5.utils.general import non_max_suppression, \ scale_coords from Y5.utils.plots import plot_one_box #np.set_printoptions(threshold=np.inf) class Density: def __init__(self, device, root): self.device = device self.root = root self.net = PKSNet().to(self.device) self.pth = "../data_MPKFNet/weigth/final.pth" # print("creat model is ", (t4 - t3) * 1000) # self.checkpointer = CheckPointer(self.net) # self.net.load_state_dict(self.checkpointer['net']) # self.checkpointer.load(self.checkpointer['net']) checkpoint = torch.load(self.pth) self.net.load_state_dict(checkpoint['net']) if gp.DebugPrint_flag: print("load model weight ok") np.set_printoptions(threshold=np.inf) def inference(self,image,imagetime,index): rootpath = os.path.join(self.root, imagetime) if not os.path.exists(rootpath): os.mkdir(rootpath) if not os.path.exists(rootpath + "/" + "Density"+"/"): os.mkdir(rootpath + "/" + "Density"+"/") target_h, target_w = image.shape[0], image.shape[1] t1 = time.time() image = cv2.resize(image, (1024, 768)) / 255.0 image = np.transpose(image, [2, 0, 1]) image = np.expand_dims(image, axis=0) image = torch.tensor(image) image = image.to(self.device) t2 = time.time() if gp.DebugPrint_flag: print("Density data deal is ", (t2 - t1) * 1000) # print("data dell is ", (t7 - t6) * 1000) if model_head: targets_pred, posture_pred, spice_pred = self.net(image) else: targets_pred = self.net(image) t3=time.time() if gp.DebugPrint_flag: print("Density inference is ", (t3 - t2) * 1000) targets_pred1 = targets_pred.cpu() num = targets_pred1.data.sum().numpy() targets_pred1 = targets_pred1.data.numpy() t4=time.time() if gp.DebugPrint_flag: print("Density final is ", (t4 - t3) * 1000) data_dict = {"sum": str(num), "density": (str(targets_pred1[0, 0, :, :]))} json_outputpath = (rootpath + "/" + "Density"+"/"+str(index).zfill(3)+".json") json_data = json.dumps(data_dict, ensure_ascii=True, indent=4) with open (json_outputpath,'w',encoding='utf-8') as f: f.write(str(json_data)) t5=time.time() if gp.DebugPrint_flag: print("Density save is ", (t5 - t4) * 1000) def inference_init(self, image): image = cv2.resize(image, (1024, 768)) / 255.0 image = np.transpose(image, [2, 0, 1]) image = np.expand_dims(image, axis=0) image = torch.tensor(image) image = image.to(self.device) if model_head: targets_pred, posture_pred, spice_pred = self.net(image) else: targets_pred = self.net(image) targets_pred1 = targets_pred.cpu() num = targets_pred1.data.sum().numpy() targets_pred1 = targets_pred1.data.numpy() if gp.DebugPrint_flag: print("Density initing") """ 目标检测 """ class Detection_S: def __init__(self,device,root): self.device = device self.root = root cfg.merge_from_file("ssd/config/hat-512.yaml") cfg.freeze() self.net =build_detection_model(cfg).to(self.device) self.checkpointer = CheckPointer(self.net, save_dir=cfg.OUTPUT_DIR) self.ckpt = "../data_MPKFNet/weigth/detection/detection.pth" self.checkpointer.load(self.ckpt, use_latest=self.ckpt is None) self.transforms = build_transforms(cfg, is_train=False) self.threshold = gp.detect_thresh self.net.eval() def inference(self,image,imagetime,index,fine_grained): with torch.no_grad(): class_names = VOCDataset.class_names rootpath = os.path.join(self.root, imagetime) if not os.path.exists(rootpath): os.mkdir(rootpath) if not os.path.exists(rootpath + "/" + "Detection" + "/"): os.mkdir(rootpath + "/" + "Detection" + "/") image_or = image height, width = image.shape[:2] t1 = time.time() image = self.transforms(image)[0].unsqueeze(0) t2 = time.time() if gp.DebugPrint_flag: print("Detection data deal is ",(t2-t1)*1000) result = self.net(image.to(self.device))[0] t3 = time.time() if gp.DebugPrint_flag: print("Detection inference is ",(t3-t2)*1000) result = result.to(torch.device("cpu")) result = result.resize((width, height)).numpy() t5 = time.time() if gp.DebugPrint_flag: print("Detection get result is ", (t5 - t3) * 1000) boxes, labels, scores = result['boxes'], result['labels'], result['scores'] indices_score = (scores > self.threshold) indices_label_left = labels <4 indices_label_right = labels >2 indices = [] for i in range(len(indices_score)): indices.append(indices_score[i] and indices_label_left[i] and indices_label_right[i]) boxes = boxes[indices] labels = labels[indices] scores = scores[indices] for i in range(len(boxes)): if boxes[i][0]<0: boxes[i][0] = 0 if boxes[i][1]<0: boxes[i][1] = 0 if boxes[i][2]<0: boxes[i][2] = 0 if boxes[i][3]<0: boxes[i][3] = 0 for i in range(len(labels)): labels[i] = 99999 boxes_fine, labels_fine, scores_fine = fine_grained.inference(image_or, boxes, labels, scores) t4 = time.time() if gp.DebugPrint_flag: print("Detection final is ", (t4 - t3) * 1000) #numb = 2 #fine_grained.inference(image_or,boxes,labels,scores) # for box in boxes: # img1 = image_or.copy()[int(box[1]):int(box[3]), int(box[0]):int(box[2]), :] # cv2.imwrite(rootpath + "/" + "Detection" + "/"+"{}.jpg".format(numb), img1) # numb+=1 #drawn_image = draw_boxes(image_or, boxes_fine, labels_fine, scores_fine).astype(np.uint8) xml_outputpath = rootpath + "/" + "Detection"+"/" + index+".xml" make_xml(str(index),image_or.shape[0],image_or.shape[1],boxes_fine,labels_fine,scores_fine,xml_outputpath,class_names) #Image.fromarray(drawn_image).save(xml_outputpath.replace(".xml",".jpg")) t5 = time.time() if gp.DebugPrint_flag: print("Detection save is ", (t5 - t4) * 1000) def inference_Red(self,image,ImagePath,fine_grained): with torch.no_grad(): class_names = VOCDataset.class_names image_or = image height, width = image.shape[:2] t1 = time.time() image = self.transforms(image)[0].unsqueeze(0) t2 = time.time() if gp.DebugPrint_flag: print("data deal is ",(t2-t1)*1000) result = self.net(image.to(self.device))[0] t3 = time.time() if gp.DebugPrint_flag: print("inference is ",(t3-t2)*1000) result = result.to(torch.device("cpu")) result = result.resize((width, height)).numpy() boxes, labels, scores = result['boxes'], result['labels'], result['scores'] indices_score = (scores > self.threshold) indices_label_left = labels <4 indices_label_right = labels >2 indices = [] for i in range(len(indices_score)): indices.append(indices_score[i] and indices_label_left[i] and indices_label_right[i]) boxes = boxes[indices] for i in range(len(boxes)): if boxes[i][0]<0: boxes[i][0] = 0 if boxes[i][1]<0: boxes[i][1] = 0 if boxes[i][2]<0: boxes[i][2] = 0 if boxes[i][3]<0: boxes[i][3] = 0 labels = labels[indices] scores = scores[indices] for i in range(len(labels)): labels[i] = 99999 boxes_fine, labels_fine, scores_fine = fine_grained.inference(image_or, boxes, labels, scores) t4 = time.time() if gp.DebugPrint_flag: print("final is ", (t4 - t3) * 1000) #numb = 2 #fine_grained.inference(image_or,boxes,labels,scores) # for box in boxes: # img1 = image_or.copy()[int(box[1]):int(box[3]), int(box[0]):int(box[2]), :] # cv2.imwrite(rootpath + "/" + "Detection" + "/"+"{}.jpg".format(numb), img1) # numb+=1 #drawn_image = draw_boxes(image_or, boxes_fine, labels_fine, scores_fine).astype(np.uint8) xml_outputpath = ImagePath.replace("/ftp/", "/ftp_Result/").replace(".JPG", ".xml").replace(".jpg",".xml") make_xml(str(ImagePath),image_or.shape[0],image_or.shape[1],boxes_fine,labels_fine,scores_fine,xml_outputpath) #Image.fromarray(drawn_image).save(xml_outputpath.replace(".xml",".jpg")) t5 = time.time() if gp.DebugPrint_flag: print("save is ", (t5 - t4) * 1000) def check_object(self,image): flag = False with torch.no_grad(): height, width = image.shape[:2] img_temp = image t1 = time.time() image = self.transforms(image)[0].unsqueeze(0) t2 = time.time() if gp.DebugPrint_flag: print("check datadeal is ",(t2-t1)*1000) result = self.net(image.to(self.device))[0] t3 = time.time() if gp.DebugPrint_flag: print("check inference is ", (t3 - t2) * 1000) result = result.to(torch.device("cpu")) t4 = time.time() if gp.DebugPrint_flag: print("check result deal is ", (t4 - t3) * 1000) result = result.resize((width, height)).numpy() boxes, labels, scores = result['boxes'], result['labels'], result['scores'] indices_score = (scores > self.threshold) indices_label_left = labels <4 indices_label_right = labels >2 indices = [] for i in range(len(indices_score)): indices.append(indices_score[i] and indices_label_left[i] and indices_label_right[i]) labels = labels[indices] boxes = boxes[indices] scores=scores[indices] if 3 in labels: flag = True if gp.DebugPrint_flag: print("boxes = ",boxes) print("check flag is ",flag) t5 = time.time() if gp.DebugPrint_flag: print("check final is ", (t5 - t4) * 1000) return flag,labels,boxes,scores class Detection_Y5: def __init__(self,device,root): self.device = device self.root = root self.img_size = 640 self.conf_thres = gp.detect_thresh self.iou_thres = 0.4 self.stride = 1 if gp.red_flag: self.weights = '../data_MPKFNet/weigth/detection/Y/best_r.pt' else: self.weights = '../data_MPKFNet/weigth/detection/Y/best.pt' # self.device = select_device(self.device) model = attempt_load(self.weights, map_location=self.device) model.to(self.device).eval() model.half() self.model = model self.names = model.module.names if hasattr( model, 'module') else model.names def preprocess(self, img): img0 = img.copy() #img = cv2.resize(img, (1024, 768)) img = letterbox(img, new_shape=self.img_size, stride=self.stride)[0] img = img[:, :, ::-1].transpose(2, 0, 1) img = np.ascontiguousarray(img) img = torch.from_numpy(img).to(self.device) img = img.half() # °ë¾«¶È img /= 255.0 # ͼÏñ¹éÒ»»¯ if img.ndimension() == 3: img = img.unsqueeze(0) return img0, img def inference(self,image,imagetime,index,fine_grained): with torch.no_grad(): rootpath = os.path.join(self.root, imagetime) if not os.path.exists(rootpath): os.mkdir(rootpath) if not os.path.exists(rootpath + "/" + "Detection" + "/"): os.mkdir(rootpath + "/" + "Detection" + "/") image_or = image img_detect = cv2.resize(image, (1024, 768)) t1 = time.time() image_scale,image = self.preprocess(img_detect) t2 = time.time() if gp.DebugPrint_flag: print("Detection data deal is ",(t2-t1)*1000) pred = self.model(image, augment=False)[0] pred = pred.float() pred = non_max_suppression(pred, self.conf_thres, self.iou_thres) t3 = time.time() if gp.DebugPrint_flag: print("Detection inference is ",(t3-t2)*1000) boxes = [] scores = [] labels = [] for det in pred: if det is not None and len(det): det[:, :4] = scale_coords( image.shape[2:], det[:, :4], image_scale.shape).round() boxes.append(det[:, :4]) scores.append(det[:, 4]) labels.append((det[:, -1])) boxes = boxes[0] scores = scores[0] labels = labels[0] t5 = time.time() if gp.DebugPrint_flag: print("Detection get result is ", (t5 - t3) * 1000) #cichuxuyaoxiugai ruguo meiyoujiance daomubiao jiuhuibaocuo if len(labels) > 0: indices_label_left = labels < 4 indices_label_right = labels > 2 indices = [] for i in range(len(indices_label_left)): indices.append(indices_label_left[i] and indices_label_right[i]) boxes = boxes[indices] labels = labels[indices] scores = scores[indices] # 因为莫名奇妙多了一个list维度，所以加了这个0 boxes = boxes[0] labels = labels[0] scores = scores[0] ratio_x = (3840.0 / 1024.0) ratio_y = (2160.0 / 768.0) for i in range(len(boxes)): boxes[i][0]=boxes[i][0] * ratio_x boxes[i][2]=boxes[i][2] * ratio_x boxes[i][1] = boxes[i][1] * ratio_y boxes[i][3] = boxes[i][3] * ratio_y if boxes[i][0]<0: boxes[i][0] = 0 if boxes[i][1]<0: boxes[i][1] = 0 if boxes[i][2]<0: boxes[i][2] = 0 if boxes[i][3]<0: boxes[i][3] = 0 for i in range(len(labels)): labels[i] = 99999 boxes_fine, labels_fine, scores_fine = fine_grained.inference(image_or, boxes, labels, scores) t4 = time.time() if gp.DebugPrint_flag: print("Detection final is ", (t4 - t3) * 1000) else: boxes_fine, labels_fine, scores_fine = [],[],[] if gp.DebugPrint_flag: print("Detection is nothong") #numb = 2 #fine_grained.inference(image_or,boxes,labels,scores) # for box in boxes: # img1 = image_or.copy()[int(box[1]):int(box[3]), int(box[0]):int(box[2]), :] # cv2.imwrite(rootpath + "/" + "Detection" + "/"+"{}.jpg".format(numb), img1) # numb+=1 #drawn_image = draw_boxes(image_or, boxes_fine, labels_fine, scores_fine).astype(np.uint8) #labels_fine = labels_fine.numpy() t6 = time.time() xml_outputpath = rootpath + "/" + "Detection"+"/" + index+".xml" make_xml(str(index),image_or.shape[0],image_or.shape[1],boxes_fine,labels_fine,scores_fine,xml_outputpath) #Image.fromarray(drawn_image).save(xml_outputpath.replace(".xml",".jpg")) t5 = time.time() if gp.DebugPrint_flag: print("Detection save is ", (t5 - t6) * 1000) def inference_Red(self,image,ImagePath,fine_grained): with torch.no_grad(): image_or = image height, width = image.shape[:2] img_detect = cv2.resize(image, (1024, 768)) t1 = time.time() image_scale, image = self.preprocess(img_detect) t2 = time.time() if gp.DebugPrint_flag: print("Detection data deal is ", (t2 - t1) * 1000) pred = self.model(image, augment=False)[0] pred = pred.float() pred = non_max_suppression(pred, self.conf_thres, self.iou_thres) t3 = time.time() if gp.DebugPrint_flag: print("Detection inference is ", (t3 - t2) * 1000) boxes = [] scores = [] labels = [] for det in pred: if det is not None and len(det): det[:, :4] = scale_coords( image.shape[2:], det[:, :4], image_scale.shape).round() boxes.append(det[:, :4]) scores.append(det[:, 4]) labels.append((det[:, -1])) boxes = boxes[0] scores = scores[0] labels = labels[0] t5 = time.time() if gp.DebugPrint_flag: print("Detection get result is ", (t5 - t3) * 1000) if len(labels) > 0: indices_label_left = labels < 4 indices_label_right = labels > 2 indices = [] for i in range(len(indices_label_left)): indices.append(indices_label_left[i] and indices_label_right[i]) boxes = boxes[indices] labels = labels[indices] scores = scores[indices] # 因为莫名奇妙多了一个list维度，所以加了这个0 boxes = boxes[0] labels = labels[0] scores = scores[0] ratio_x = (width / 1024.0) ratio_y = (height / 768.0) for i in range(len(boxes)): boxes[i][0] = boxes[i][0] * ratio_x boxes[i][2] = boxes[i][2] * ratio_x boxes[i][1] = boxes[i][1] * ratio_y boxes[i][3] = boxes[i][3] * ratio_y if boxes[i][0] < 0: boxes[i][0] = 0 if boxes[i][0] > 3840: boxes[i][0] = 3839 if boxes[i][1] < 0: boxes[i][1] = 0 if boxes[i][1] > 2160: boxes[i][1] = 2159 if boxes[i][2] < 0: boxes[i][2] = 0 if boxes[i][2]> 3840: boxes[i][2] = 3839 if boxes[i][3] < 0: boxes[i][3] = 0 if boxes[i][3] > 2160: boxes[i][3] = 2159 for i in range(len(labels)): labels[i] = 99999 boxes_fine, labels_fine, scores_fine = fine_grained.inference(image_or, boxes, labels, scores) t4 = time.time() if gp.DebugPrint_flag: print("Detection final is ", (t4 - t3) * 1000) t6 = time.time() xml_outputpath = ImagePath.replace("/ftp/", "/ftp_Result/").replace(".JPG", ".xml").replace(".jpg", ".xml") make_xml(str(ImagePath), image_or.shape[0], image_or.shape[1], boxes_fine, labels_fine, scores_fine, xml_outputpath) # Image.fromarray(drawn_image).save(xml_outputpath.replace(".xml",".jpg")) t5 = time.time() if gp.DebugPrint_flag: print("Detection save is ", (t5 - t6) * 1000) else: boxes_fine, labels_fine, scores_fine = [], [], [] if gp.DebugPrint_flag: print("Detection is nothong") def check_object(self,img_o): flag = False with torch.no_grad(): # path = '/data/Hatcher/ouput1/0001_20210414151147/RawImage/0099.jpg' # img = cv2.imread(path) # img = cv2.resize() height, width = img_o.shape[:2] img = cv2.resize(img_o,(1024,768)) t1 = time.time() image_scale, image = self.preprocess(img) t2 = time.time() if gp.DebugPrint_flag: print("Check data deal is ", (t2 - t1) * 1000) pred = self.model(image, augment=False)[0] pred = pred.float() pred = non_max_suppression(pred, self.conf_thres, self.iou_thres) t3 = time.time() if gp.DebugPrint_flag: print("Check inference is ", (t3 - t2) * 1000) boxes = [] scores = [] labels = [] for det in pred: if det is not None and len(det): det[:, :4] = scale_coords( image.shape[2:], det[:, :4], image_scale.shape).round() boxes.append(det[:, :4]) scores.append(det[:, 4]) labels.append((det[:, -1])) boxes = boxes[0] scores = scores[0] labels = labels[0] t5 = time.time() if gp.DebugPrint_flag: print("Check get result is ", (t5 - t3) * 1000) if len(labels) > 0: indices_label_left = labels < 4 indices_label_right = labels > 2 indices = [] for i in range(len(indices_label_left)): indices.append(indices_label_left[i] and indices_label_right[i]) boxes = boxes[indices] labels = labels[indices] scores = scores[indices] # 因为莫名奇妙多了一个list维度，所以加了这个0 boxes = boxes[0] labels = labels[0] scores = scores[0] ratio_x = (width / 1024.0) ratio_y = (height / 768.0) for i in range(len(boxes)): boxes[i][0] = boxes[i][0] * ratio_x boxes[i][2] = boxes[i][2] * ratio_x boxes[i][1] = boxes[i][1] * ratio_y boxes[i][3] = boxes[i][3] * ratio_y if boxes[i][0] < 0: boxes[i][0] = 0 if boxes[i][0] > 3840: boxes[i][0] = 3839 if boxes[i][1] < 0: boxes[i][1] = 0 if boxes[i][1] > 2160: boxes[i][1] = 2159 if boxes[i][2] < 0: boxes[i][2] = 0 if boxes[i][2] > 3840: boxes[i][2] = 3839 if boxes[i][3] < 0: boxes[i][3] = 0 if boxes[i][3] > 2160: boxes[i][3] = 2159 else: pass if 3 in labels: flag = True if gp.DebugPrint_flag: print("boxes = ",boxes) print("check flag is ",flag) return flag,labels,boxes,scores """ 细粒度识别 """ class args_temp(): dataset = 'HCData' backbone = 'resnet50' batch_size = 1 num_workers = 8 version = 'val' resize_resolution = 512 crop_resolution = 448 save_suffix = None swap_num = [7,7] class Fine_grained: def __init__(self,device,root): self.device = device self.root = root self.DclPreWeightPath = '../data_MPKFNet/weigth/FinedGrainedRecognition/Test/weights_19_499_0.9998_1.0000.pth' # parser = argparse.ArgumentParser(description='dcl parameters') # parser.add_argument('--data', dest='dataset', default='HCData', type=str) # parser.add_argument('--backbone', dest='backbone', default='resnet50', type=str) # parser.add_argument('--b', dest='batch_size', default=1, type=int) # parser.add_argument('--nw', dest='num_workers', default=8, type=int) # parser.add_argument('--ver', dest='version', default='val', type=str) # parser.add_argument('--size', dest='resize_resolution', default=512, type=int) # parser.add_argument('--crop', dest='crop_resolution', default=448, type=int) # parser.add_argument('--ss', dest='save_suffix', default=None, type=str) # parser.add_argument('--acc_report', dest='acc_report', action='store_true') # parser.add_argument('--swap_num', default=[7, 7], nargs=2, metavar=('swap1', 'swap2'), type=int, # help='specify a range') self.args = args_temp() self.Config = LoadConfig(self.args, 's') self.transformers = load_data_transformers(self.args.resize_resolution, self.args.crop_resolution, self.args.swap_num) self.net = MainModel(self.Config).to(device) model_dict = self.net.state_dict() pretrained_dict = torch.load(self.DclPreWeightPath) model_dict.update(pretrained_dict) self.net.load_state_dict(model_dict) if gp.DebugPrint_flag: print("load model weight is ok") self.net.train(False) def inference(self,img,imagelist,labellist,scorelist): with torch.no_grad(): for i,box in enumerate(imagelist): Img = img.copy()[int(box[1]):int(box[3]), int(box[0]):int(box[2]), :] if (int(box[3])-int(box[1])>100) and (int(box[2])-int(box[0])>100): t1 = time.time() Img = Image.fromarray(cv2.cvtColor(Img, cv2.COLOR_BGR2RGB)) Img = self.transformers['test_totensor'](Img) Img = Img.unsqueeze(0) inputs = Variable(Img.cuda()) t2 = time.time() if gp.DebugPrint_flag: print("fine data deal is ", (t2 - t1) * 1000) outputs = self.net(inputs) t3 = time.time() if gp.DebugPrint_flag: print("fine inference is ", (t3 - t2) * 1000) top3_val, top3_pos = torch.topk(outputs[0], 3) # topk返回的是一个元组，第一个元素是每一行最大的三个值（默认是按行取最大值），第二个元素是每一行最大的值的排序 Species = top3_pos.cpu().numpy()[0][0] ConfidenceLevel = top3_val.cpu().numpy()[0][0] ConfidenceLevel = ConfidenceLevel/26.0 if ConfidenceLevel >1: ConfidenceLevel = 1.0* random.uniform(0, 1) if gp.DebugPrint_flag: print("hhhhhhhhhhhhhhhhhhhhhhh is ",ConfidenceLevel) labellist[i] = Species scorelist[i] = ConfidenceLevel t4 = time.time() if gp.DebugPrint_flag: print("fine data final is ", (t4 - t3) * 1000) return imagelist, labellist, scorelist class Check_start (threading.Thread): def __init__(self, threadID, name, img,DetectionClass): threading.Thread.__init__(self) self.threadID = threadID self.name = name self.img = img self.DetectionClass = DetectionClass def run(self): if gp.DebugPrint_flag: print(self.name + " is start") gp.is_ex_bird ,labels, boxes ,scores= self.DetectionClass.check_object(self.img) if gp.is_ex_bird : if gp.DebugPrint_flag: print(gp.is_ex_bird) gp.image_time = gp.Camera_device +time.strftime('%Y%m%d%H%M%S',time.localtime(time.time())) if gp.DebugPrint_flag: print(self.name + " is finish") class Check_start_fordataselect (threading.Thread): def __init__(self, threadID, name, img,DetectionClass,fine_grained): threading.Thread.__init__(self) self.threadID = threadID self.name = name self.img = img self.DetectionClass = DetectionClass self.fine_grained = fine_grained def run(self): if gp.DebugPrint_flag: print(self.name + " is start") is_ex_bird ,labels, boxes ,scores= self.DetectionClass.check_object(self.img) image_or = self.img.copy() if is_ex_bird : for i in range(len(boxes)): if boxes[i][0] < 0: boxes[i][0] = 0 if boxes[i][1] < 0: boxes[i][1] = 0 if boxes[i][2] < 0: boxes[i][2] = 0 if boxes[i][3] < 0: boxes[i][3] = 0 for i in range(len(labels)): labels[i] = 99999 boxes_fine, labels_fine, scores_fine = self.fine_grained.inference(self.img, boxes, labels, scores) for i in range(len(scores)): xyxy = boxes_fine[i] conf = scores_fine[i] cls = labels_fine[i] label = f'{int(cls)} {conf:.2f}' palette = (2 ** 11 - 1, 2 ** 15 - 1, 2 ** 20 - 1) colors = [int((p * (cls ** 2 - cls + 1)) % 255) for p in palette] plot_one_box(xyxy, self.img, label=label, color=colors, line_thickness=1) cv2.imwrite("../data_MPKFNet/RTSP/" + "{}.jpg".format(time.strftime('%Y%m%d%H%M%S', time.localtime(time.time()))), image_or) cv2.imwrite( "../data_MPKFNet/RTSP/result/" + "{}.jpg".format(time.strftime('%Y%m%d%H%M%S', time.localtime(time.time()))), self.img) if gp.DebugPrint_flag: print(self.name + " is finish") class Density_mul (threading.Thread): def __init__(self, threadID, name,imagetime,DensityClass): threading.Thread.__init__(self) self.threadID = threadID self.name = name self.imagetime = imagetime self.DensityClass = DensityClass def run(self): gp.Density_thread_flag = True if gp.DebugPrint_flag: print(self.name +self.imagetime," is starting") i = 0 while (i<=29): # if(len(gp.ImgDealQueue)<=i): # gp.event.wait() # if ((i+1)%25==0): #说明是第25*n帧,间隔1s处理一次 if gp.DebugPrint_flag: print("Density is1 ", len(gp.ImgDealQueue), " ", i) print("Density is waiting") if (len(gp.ImgDealQueue) <= i): gp.eventDensity.wait() if gp.DebugPrint_flag: print("Density is2 ", len(gp.ImgDealQueue), " ", i) print("Density is not waiting") self.DensityClass.inference(gp.ImgDealQueue[i],self.imagetime,str(i).zfill(4)) gp.eventDensity.clear() i += 1 gp.Density_thread_flag = False gp.is_ex_bird = False gp.destory_image = True if gp.DebugPrint_flag: print(self.name + self.imagetime," is over") class Detecntion_mul (threading.Thread): def __init__(self, threadID, name,imagetime,DetectionClass,Fine_grainedClass): threading.Thread.__init__(self) self.threadID = threadID self.name = name self.imagetime = imagetime self.DetectionClass = DetectionClass self.Fine_grainedClass = Fine_grainedClass def run(self): gp.Detection_thread_flag = True if gp.DebugPrint_flag: print(self.name ,self.imagetime," is starting") i = 0 while (i<=29): if gp.DebugPrint_flag: print("Detection is1 ",len(gp.ImgDealQueue)," ",i) # if (len(gp.ImgDealQueue) <= i): # gp.event.wait() # if ((i+1)%25==0): #说明是第25*n帧,间隔1s处理一次 print("Detection is waiting") if (len(gp.ImgDealQueue) <= i): gp.eventDetection.wait() if gp.DebugPrint_flag: print("Detection is2 ", len(gp.ImgDealQueue), " ", i) print("Detection is not waiting") self.DetectionClass.inference(gp.ImgDealQueue[i],self.imagetime,str(i).zfill(4),self.Fine_grainedClass) gp.eventDetection.clear() i+=1 gp.Detection_thread_flag = False gp.is_ex_bird = False gp.destory_image = True if gp.DebugPrint_flag: print(self.name ,self.imagetime," is over") class SaveImage_mul (threading.Thread): def __init__(self, threadID, name,imagetime,root): threading.Thread.__init__(self) self.threadID = threadID self.name = name self.imagetime = imagetime self.root = root self.rootpath = os.path.join(self.root, self.imagetime) def run(self): gp.SaveImage_thread_flag = True if not os.path.exists(self.rootpath): os.mkdir(self.rootpath) if not os.path.exists(self.rootpath + "/" + "RawImage" + "/"): os.mkdir(self.rootpath + "/" + "RawImage" + "/") if gp.DebugPrint_flag: print(self.name ,self.imagetime," is starting") i = 0 while (i<=29): # if(len(gp.ImgDealQueue)<=i): # gp.event.wait() # if ((i+1)%25==0): #说明是第25*n帧,间隔1s处理一次 if gp.DebugPrint_flag: print("Saveimage is1 ", len(gp.ImgDealQueue), " ", i) print("Saveimage is waiting") if (len(gp.ImgDealQueue) <= i): gp.eventSaveimage.wait() if gp.DebugPrint_flag: print("Saveimage is2 ", len(gp.ImgDealQueue), " ", i) print("Saveimage is not waiting") cv2.imwrite(os.path.join(self.rootpath,"RawImage","{}.jpg".format(str(i).zfill(4))),gp.ImgDealQueue[i]) gp.eventSaveimage.clear() if gp.DebugPrint_flag: print("Saveimage finish once") i+=1 gp.SaveImage_thread_flag = False gp.is_ex_bird = False gp.destory_image = True if gp.DebugPrint_flag: print(self.name ,self.imagetime," is over") class ThreadMonitor_mul (threading.Thread): def __init__(self, threadID, name): threading.Thread.__init__(self) self.threadID = threadID self.name = name def run(self): if gp.DebugPrint_flag: print(self.name," is starting") while 1: # print(gp.Density_thread_flag,gp.Detection_thread_flag,gp.SaveImage_thread_flag) if (gp.Density_thread_flag) == False and (gp.Detection_thread_flag == False) and (gp.SaveImage_thread_flag == False): #说明三个进程都已经结束了 gp.ImgDealQueue = [] class isopenmul (threading.Thread): def __init__(self, threadID, name,DetectionClass): threading.Thread.__init__(self) self.threadID = threadID self.name = name self.DetectionClass = DetectionClass def run(self): while 1: # print(gp.Density_thread_flag,gp.Detection_thread_flag,gp.SaveImage_thread_flag) if (gp.Vedio_thread_flag == False) and (gp.Density_thread_flag == False) and ( gp.Detection_thread_flag == False) and (gp.SaveImage_thread_flag == False) and ( gp.is_ex_bird == False) and gp.num_detection > 24: gp.num_detection = 0 if gp.DebugPrint_flag: print( "check is start") gp.is_ex_bird, labels, boxes, scores = self.DetectionClass.check_object(gp.frame_openmul) if gp.is_ex_bird: if gp.DebugPrint_flag: print(gp.is_ex_bird) gp.image_time = gp.Camera_device + time.strftime('%Y%m%d%H%M%S', time.localtime(time.time())) if gp.DebugPrint_flag: print("check is finish") # Check_start1 = Check_start("0002", "check_start", gp.frame_openmul, self.DetectionClass) # Check_start1.start() if __name__ == '__main__': t1 = time.time() image = cv2.imread("/home/wanghc/output1/0001_20210413155842/RawImage/0174.jpg") t2 = time.time() camera_num = "0001" image_time = time.strftime('%Y%m%d%H%M%S',time.localtime(time.time())) image_time = camera_num + image_time root = "/home/wanghc/output" Density1 = Density(gp.devices,root) Detection1 = Detection_Y5(gp.devices,root) Fine_grained1 = Fine_grained(gp.devices,root) os.environ['CUDA_VISIBLE_DEVICES'] = '4' device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu') path = '/home/wanghc/output1/001020210509163621/RawImage/0000.jpg' img = cv2.imread(path) # BGR Detection1.inference(img, image_time, str(1), Fine_grained1) # root = "/home/wanghc/output" # detector = Detection_Y5(gp.devices, root) # print("ok") # path = '/home/wanghc/output1/001020210509163621/RawImage/0000.jpg' # img = cv2.imread(path) # BGR # img_show = img#cv2.resize(img, (1024, 768)) # names = detector.model.module.names if hasattr(detector.model, 'module') else detector.model.names # colors = [[random.randint(0, 255) for _ in range(3)] for _ in names] # flag, labels,pred_boxes, scores = detector.check_object(img_show) # print(len(scores)) # #img_show = cv2.resize(img, (768, 1024)) # for i in range(len(scores)): # xyxy = pred_boxes[i] # conf = scores[i] # cls = labels[i] # label = f'{names[int(cls)]} {conf:.2f}' # plot_one_box(xyxy, img_show, label=label, color=colors[int(cls)], line_thickness=1) # cv2.imshow(str(path), img_show) # t1 = time.time() # cv2.imwrite("./a.jpg", img_show) # t2 = time.time() # print((t2 - t1) * 1000) # cv2.waitKey(-1) # while(1): # pass