0822更新代码逻辑，更新图片分类代码

Bank_second_part/model_annotation/create_xml.py

@@ -2,7 +2,7 @@
 from lxml.etree import Element, SubElement, tostring
 
 
-def create_xml(boxs,img_shape,xml_path):
+def create_xml(boxs, img_shape, xml_path):
     """
     创建xml文件，依次写入xml文件必备关键字
     :param boxs:   txt文件中的box
@@ -22,7 +22,7 @@ def create_xml(boxs,img_shape,xml_path):
     node_depth = SubElement(node_size, 'depth')
     node_depth.text = str(img_shape[2])
 
-    if len(boxs)>=1:        # 循环写入box
+    if len(boxs) >= 1:        # 循环写入box
         for box in boxs:
             node_object = SubElement(node_root, 'object')
             node_name = SubElement(node_object, 'name')
@@ -45,14 +45,9 @@ def create_xml(boxs,img_shape,xml_path):
 
     xml = tostring(node_root, pretty_print=True)   # 格式化显示，该换行的换行
 
-    file_name = img_shape[3].split(".")[0]
+    # file_name = img_shape[3].split(".")[0]
     # filename = xml_path+"/{}.xml".format(file_name)
 
     f = open(xml_path, "wb")
     f.write(xml)
     f.close()
- 
-
-
-
-

Bank_second_part/model_annotation/holisticDet.py

@@ -51,12 +51,7 @@ class MediapipeProcess:
 
     result_dict = {'hand_bbox':[right_hand_bbox,left_hand_bbox]}
 
-    # print('hand_bbox:',result_dict)
-
-
     return result_dict
-        
-
 
   def point_to_bbox(result_list):
       
@@ -78,29 +73,7 @@ class MediapipeProcess:
 
       bbox_list = left_top + right_bottom
 
-      # print('bbox:',bbox)
-      # print('bbox_list:',bbox_list)
-
-
-      # bbox_list = []
-
-      # bbox_list.append(bbox[0][0])
-      # bbox_list.append(bbox[0][1])
-      # bbox_list.append(bbox[2][0])
-      # bbox_list.append(bbox[2][1])
  
       return bbox_list
 
-    else:
-      pass 
-
-
-
-
-
-   
-
 
-# if __name__ == '__main__':  
-#   # media_holistic(video_file='E:/Bank_files/Bank_02/dataset/video_person/after_1/0711-1_199_0.avi',
-  #                video_save_path='E:/Bank_files/Bank_02/videos_mediapipe/test_data/0725_test')

Bank_second_part/model_annotation/model_annotations_process.py

@@ -12,142 +12,148 @@ import queue
 
 class DetProcess():
 
-    def __init__(self,person_det_model,hand_det_model):
+    def __init__(self, person_det_model, hand_det_model):
 
         self.person_det_model = person_det_model
         self.hand_det_model = hand_det_model
 
-        
-    def get_person_cut(self,frame,det_dict,imgsize):
+    def get_person_cut(self, frame, det_dict, imgsize):
 
-        # person_det_dict = [perdict for perdict in det_dict if list(perdict.keys())[0] == 'person']
-
-        # print('person_det_dict:',det_dict)
         person_list = tools_function.get_dict_values(det_dict)
 
         # 坐标参数修正
-        person_bbox_list = tools_function.para_list_correction(images_size=imgsize,bbox_list=person_list,dertpara=5)
+        person_bbox_list = tools_function.para_list_correction(
+            images_size=imgsize, bbox_list=person_list, dertpara=5)
 
         frame_list = []
         for per_bbox in person_bbox_list:
 
             # 裁剪后人的图片
-            person_cut_frame = tools_function.img_cut(frame=frame,bbox=per_bbox)
+            person_cut_frame = tools_function.img_cut(
+                frame=frame, bbox=per_bbox)
 
-            frame_cut_dict = {tuple(per_bbox):person_cut_frame}
+            frame_cut_dict = {tuple(per_bbox): person_cut_frame}
 
             frame_list.append(frame_cut_dict)
-        
+
         return frame_list
-        
-    def hand_det(self,person_cut_frame,per_bbox):
-    
-        # print('11111')
-        hands_result = MediapipeProcess.mediapipe_det(image=person_cut_frame,holistic=self.hand_det_model)
-        hands_result_dict = MediapipeProcess.get_analysis_result(image=person_cut_frame,results=hands_result)
+
+    def hand_det(self, person_cut_frame, per_bbox):
+
+        hands_result = MediapipeProcess.mediapipe_det(
+            image=person_cut_frame, holistic=self.hand_det_model)
+        hands_result_dict = MediapipeProcess.get_analysis_result(
+            image=person_cut_frame, results=hands_result)
         hands_list = tools_function.select_list(hands_result_dict['hand_bbox'])
 
-        # print('hands_list:',hands_list)
         imgsize2 = person_cut_frame.shape
 
         # 手部坐标修正
-        hands_bbox_list = tools_function.para_list_correction(images_size=imgsize2,bbox_list=hands_list,dertpara=5)
-
-        # print('hands_bbox_list:',hands_bbox_list)
+        hands_bbox_list = tools_function.para_list_correction(
+            images_size=imgsize2, bbox_list=hands_list, dertpara=5)
 
         hand_bbox_list = []
         for hand in hands_bbox_list:
 
-            hands_result_list = tools_function.change_bbox(bbox_person=[per_bbox[0],per_bbox[1]],bbox_hand=hand)
-
-            # print('hands_result_list:',hands_result_list)
+            hands_result_list = tools_function.change_bbox(
+                bbox_person=[per_bbox[0], per_bbox[1]], bbox_hand=hand)
 
-            re_dict = {'hands':hands_result_list}
+            re_dict = {'hands': hands_result_list}
 
             hand_bbox_list.append(re_dict)
 
-        # hands_result_original_dict = {'results':hand_bbox_list}
-
-        # print(hands_result_original_dict)
-
         return hand_bbox_list
 
-    def save_annotations_xml(self,xml_save_file,save_infors,images):
+    def save_annotations_xml(self, xml_save_file, save_infors, images):
 
-        # images = save_infors['images']
         results = save_infors
-        img  = os.path.basename(images)
+        img = os.path.basename(images)
         img_frame = cv2.imread(images)
-        xml_save_path = os.path.join(xml_save_file,img.split('.')[0] + '.xml')
-        w,h,d = img_frame.shape
-        img_shape = (w,h,d,img)
+        xml_save_path = os.path.join(xml_save_file, img.split('.')[0] + '.xml')
+        w, h, d = img_frame.shape
+        img_shape = (w, h, d, img)
 
         if os.path.isfile(xml_save_path):
 
-            add_labels = add_xml(inforsDict=results,xmlFilePath=xml_save_path)
+            add_labels = add_xml(inforsDict=results, xmlFilePath=xml_save_path)
         else:
-            create_new = create_xml(boxs=results,img_shape=img_shape,xml_path=xml_save_path)
-
-
-    def person_cut_process(self,images,img_save_files):
+            create_new = create_xml(
+                boxs=results, img_shape=img_shape, xml_path=xml_save_path)
 
+    def person_cut_process(self, images, img_save_files):
+        '''
+        使用模型将检测到的人从大图中裁剪出来
+        images：图片全路径
+        img_save_files：裁剪后保存小图的文件夹
+        通过设置labels_name_list列表中的标签名，调整需要裁剪出来的目标
+        '''
 
         frame = cv2.imread(images)
         imgsize = frame.shape
         labels_name_list = ['person']
 
         per_det_dict = analysis_yolov8(frame=frame,
-                                        model_coco=self.person_det_model,
-                                        labels_names=labels_name_list,
-                                        confidence_set=0.2)
-        
-        per_frame_cut = self.get_person_cut(frame=frame,det_dict=per_det_dict,imgsize=imgsize)
-        per_frame_list = [value for dictionary in per_frame_cut for value in dictionary.values()]
+                                       model_coco=self.person_det_model,
+                                       labels_names=labels_name_list,
+                                       confidence_set=0.2)
+
+        per_frame_cut = self.get_person_cut(
+            frame=frame, det_dict=per_det_dict, imgsize=imgsize)
+        per_frame_list = [
+            value for dictionary in per_frame_cut for value in dictionary.values()]
 
-        # print('per_frame_list:',per_frame_list)
-      
-        for id_num,cut_frame in enumerate(per_frame_list):
+        for id_num, cut_frame in enumerate(per_frame_list):
 
-            cut_frame_save = tools_function.img_write(frame=cut_frame,img_file=images,id_num=id_num,save_file=img_save_files)
-    
-    def hands_det_process(self,images,xml_save_file):
+            cut_frame_save = tools_function.img_write(
+                frame=cut_frame, img_file=images, id_num=id_num, save_file=img_save_files)
+
+    def hands_det_process(self, images, xml_save_file):
+        '''
+        使用目标检测模型检测到行人，然后串联使用mediapipe模型检测到人的手部，后将检测到的手部坐标框保存成xml标注文件
+        images：检测图片的全路径
+        xml_save_file: 保存标注文件路径
+        '''
 
         frame = cv2.imread(images)
         imgsize = frame.shape
         labels_name_list = ['person']
 
         per_det_dict = analysis_yolov8(frame=frame,
-                                        model_coco=self.person_det_model,
-                                        labels_names=labels_name_list,
-                                        confidence_set=0.2)
-        
-        per_frame_cut = self.get_person_cut(frame=frame,det_dict=per_det_dict,imgsize=imgsize)
+                                       model_coco=self.person_det_model,
+                                       labels_names=labels_name_list,
+                                       confidence_set=0.2)
+
+        per_frame_cut = self.get_person_cut(
+            frame=frame, det_dict=per_det_dict, imgsize=imgsize)
 
         for frame_dict in per_frame_cut:
 
             per_bbox = list(frame_dict.keys())[0]
             person_cut_frame = list(frame_dict.values())[0]
 
-            hands_dict = self.hand_det(person_cut_frame=person_cut_frame,per_bbox=per_bbox)
-
-            self.save_annotations_xml(xml_save_file=xml_save_file,save_infors=hands_dict,images=images)
+            hands_dict = self.hand_det(
+                person_cut_frame=person_cut_frame, per_bbox=per_bbox)
 
+            self.save_annotations_xml(
+                xml_save_file=xml_save_file, save_infors=hands_dict, images=images)
 
-        # person_det = self.detect_yolo(images_path=images,labels_name_list=labels_name_list)
-
-    def det_process(self,images,xml_save_file):
+    def det_process(self, images, xml_save_file):
+        '''
+        检测指定目标，并保存检测结果到xml文件中
+        通过设置labels_name_list列表中的标签名，调整需要裁剪出来的目标
+        '''
 
         frame = cv2.imread(images)
         imgsize = frame.shape
-        labels_name_list = ['cell phone','mouse','keyboard']
+        labels_name_list = ['cell phone', 'mouse', 'keyboard']
 
         per_det_dict = analysis_yolov8(frame=frame,
-                                        model_coco=self.person_det_model,
-                                        labels_names=labels_name_list,
-                                        confidence_set=0.2)
-               
-        self.save_annotations_xml(xml_save_file=xml_save_file,save_infors=per_det_dict,images=images)
+                                       model_coco=self.person_det_model,
+                                       labels_names=labels_name_list,
+                                       confidence_set=0.2)
 
+        self.save_annotations_xml(
+            xml_save_file=xml_save_file, save_infors=per_det_dict, images=images)
 
 
 if __name__ == '__main__':
@@ -157,25 +163,18 @@ if __name__ == '__main__':
     img_save_files = 'images_cut'
     xml_save_file = 'annotations'
 
-
     # 初始化目标检测
     person_model = YOLO("model_files/yolov8x.pt")
 
     # 初始化mediapipe
     mp_holistic = mp.solutions.holistic
-    holistic = mp_holistic.Holistic(min_detection_confidence=0.1,min_tracking_confidence=0.1)
+    holistic = mp_holistic.Holistic(
+        min_detection_confidence=0.1, min_tracking_confidence=0.1)
 
-    Det = DetProcess(person_det_model=person_model,hand_det_model=holistic)
+    Det = DetProcess(person_det_model=person_model, hand_det_model=holistic)
 
     for images in images_list:
 
-        # Det.person_cut_process(images=images,img_save_files=img_save_files)
-        Det.hands_det_process(images=images,xml_save_file=xml_save_file)
-        Det.det_process(images=images,xml_save_file=xml_save_file)
-
-                                
-                
-
-
-
-
+        Det.person_cut_process(images=images, img_save_files=img_save_files)
+        Det.hands_det_process(images=images, xml_save_file=xml_save_file)
+        Det.det_process(images=images, xml_save_file=xml_save_file)

Bank_second_part/model_annotation/personDet.py

@@ -1,43 +1,40 @@
 
-def analysis_yolov8(frame, model_coco,labels_names,confidence_set):
+def analysis_yolov8(frame, model_coco, labels_names, confidence_set):
 
     # 第一步：用COCO数据集推理
     results_coco = model_coco(frame)
 
     re_list = []
-    
+
     if results_coco:
 
         for r in results_coco:
 
             boxes = r.boxes
-            
+
             idx = 0
 
             for box in boxes:
 
                 idx += 1
-                b = box.xyxy[0]  # get box coordinates in (top, left, bottom, right) format
+                # get box coordinates in (top, left, bottom, right) format
+                b = box.xyxy[0]
                 c = box.cls
 
                 # 保存标签和坐标值作为返回结果
                 blist = b.tolist()
                 labels_name = model_coco.names[int(c)]
-
-                # print('labels_name:',labels_name)
-    
                 confidence = float(box.conf)
-                        
                 confidence = round(confidence, 2)
 
                 # 过滤置信度以下目标
                 if confidence < confidence_set:
 
                     continue
-                
+
                 if labels_name in labels_names:
-                # 一个结果字典
-                    re_dict = {labels_name:blist}
+                    # 一个结果字典
+                    re_dict = {labels_name: blist}
 
                     re_list.append(re_dict)
 

Bank_second_part/model_annotation/select-picture-hand.py

@@ -0,0 +1,48 @@
+import os
+import cv2
+from tqdm import tqdm
+
+#总数据集图像路径
+imgway_1="E:/data-temporary/zhongtian_nianhou/shoes/del"
+#符合要求图像的路径
+imgway_2="E:/data-temporary/zhongtian_nianhou/shoes/0216_dataset/background"
+#不符合要求图像的路径.
+imgway_3="G:/zhongtian_second/hangche/images background"
+#误触
+imgway_7='G:/zhongtian_second/hangche/images background'
+
+
+
+for filename in tqdm(os.listdir(r"%s" % imgway_1)):
+    
+
+    img = cv2.imread("%s/%s"%(imgway_1,filename))
+    h, w, c = img.shape
+    resize_width = int(w * 2)
+    resize_height = int(h * 2)
+    # cv2.namedWindow(img, cv2.WINDOW_NORMAL)
+    # cv2.resizeWindow(img, resize_width, resize_height) #缩小显示
+    img = cv2.resize(img,(resize_width,resize_height)) #更改图片大小，设定为指定大小o
+    # # h = 640
+    # # w = 300
+    # cv2.resizeWindow(img, resize_height, resize_width)
+    # cv2.namedWindow('datasets',0)
+    cv2.imshow('datasets', img)
+
+    k=cv2.waitKey(0)& 0xFF
+    if k == ord('o'):
+        cv2.imwrite('%s/%s' % (imgway_2, filename),img)
+        print('%s符合要求!' % filename)
+        cv2.destroyAllWindows()
+
+    elif k == ord('k'):
+        pass
+    #         cv2.imwrite('%s/%s' % (imgway_3, filename),img)
+    #         print('%s不符合要求！' % filename)
+    #         cv2.destroyAllWindows()
+
+    else:
+        cv2.imwrite('%s/%s' % (imgway_7, filename),img)
+        print('误触错误的键！背景图%s保存至mistake文件夹！' % filename)
+        cv2.destroyAllWindows()
+        

Bank_second_part/model_annotation/tools_function.py

@@ -78,8 +78,6 @@ def select_list(result_list):
             if result == None:
                 pass
             else:
-
-                # result_bbox = select_bbox(result)
                 result_only.append(result)
 
         return result_only
@@ -87,7 +85,6 @@ def select_list(result_list):
 
 def para_list_correction(images_size, bbox_list, dertpara):
 
-    # bbox_list_iou = del_bbox(bbox_list)
 
     updata_result_list = []