【Demo 1】基于object_detection API的行人检测 3：模型训练并在OpenCV调用模型

训练准备

模型选择

选择ssd_mobilenet_v2_coco模型，下载地址（https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md），解压到./Pedestrian_Detection/ssd_mobilenet_v2_coco_2018_03_29.

修改object_detection配置文件

进入目录./Pedestrian_Detection/models/research/object_detection/samples/configs 找到对应的模型配置文件ssd_mobilenet_v2_coco.config修改配置文件。

根据提示信息：

1、第9行，检测类别把90改为1，因为我们只检测行人，只有一个类别。

2、修改除提示外所有的

　　2.1、第一个（156行）是我们所需模型的路径，即上一步下载好的：./Pedestrian_Detection/ssd_mobilenet_v2_coco_2018_03_29/model.ckpt

　　2.2、第二个（175行）是train.record文件的路径，上一篇中我们准备好的record文件：./Pedestrian_Detection/project/pedestrian_train/data/pascal_train.record

　　2.3、第三个（177行）是上一篇准备好的label_map.pbtxt的路径：./Pedestrian_Detection/project/pedestrian_train/data/label_map.pbtxt

　　2.4、第四个（189行）是eval.record文件的路径，上一篇中我们准备好的record文件：./Pedestrian_Detection/project/pedestrian_train/data/pascal_eval.record

　　2.5、第五个（191行）同2.3

这样config文件就修改完成了。然后把它放到：./Pedestrian_Detection/project/pedestrian_train/models目录下。最后在该目录下创建两个文件夹：train 和 eval，用于储存训练和验证的记录。

开始训练

打开命令行窗口

在research目录下输入：

(dl) D:\Study\dl\Pedestrian_Detection\models\research>python object_detection/legacy/train.py --train_dir=D:\Study\dl\Pedestrian_Detection\project\pedestrian_train\models\train --pipeline_config_path=D:\Study\dl\Pedestrian_Detection\project\pedestrian_train\models\ssd_mobilenet_v2_coco.config --logtostderr

即可开始训练。

这里我们选择2000次之后，按ctrl+c结束训练。训练的详细信息可通过tensorboard来进行查看（这里不再赘述）。

查看我们的训练记录：

导出模型

这里我们选择第2391次的训练数据来生成模型。

把下图4个文件放到：./Pedestrian_Detection/pedestrian_data/model  目录下

在命令行窗口下输入命令：

(dl) D:\Study\dl\Pedestrian_Detection\models\research>python object_detection/export_inference_graph.py --input_type=image_tensor --pipeline_config_path=D:\Study\dl\Pedestrian_Detection\project\pedestrian_train\models\ssd_mobilenet_v2_coco.config --trained_checkpoint_prefix=D:\Study\dl\Pedestrian_Detection\pedestrian_data\model\model.ckpt-2391 --output_directory=D:\Study\dl\Pedestrian_Detection\pedestrian_data\test

查看发现对应的目录下已经生成了一系列的模型文件：

测试模型

测试代码：

 import os
 import sys

 import cv2
 import numpy as np
 import tensorflow as tf

 sys.path.append("..")
 from object_detection.utils import label_map_util
 from object_detection.utils import visualization_utils as vis_util

 ##################################################

 ##################################################

 # Path to frozen detection graph
 PATH_TO_CKPT = 'D:/Study/dl/Pedestrian_Detection/pedestrian_data/test/frozen_inference_graph.pb'

 # List of the strings that is used to add correct label for each box.
 PATH_TO_LABELS = os.path.join('D:/Study/dl/Pedestrian_Detection/project/pedestrian_train/data', 'label_map.pbtxt')

 NUM_CLASSES = 1
 detection_graph = tf.Graph()
 with detection_graph.as_default():
     od_graph_def = tf.GraphDef()
     with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid:
         serialized_graph = fid.read()
         od_graph_def.ParseFromString(serialized_graph)
         tf.import_graph_def(od_graph_def, name='')

 label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
 categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True)
 category_index = label_map_util.create_category_index(categories)

 def load_image_into_numpy_array(image):
     (im_width, im_height) = image.size
     return np.array(image.getdata()).reshape(
       (im_height, im_width, 3)).astype(np.uint8)

 with detection_graph.as_default():
     with tf.Session(graph=detection_graph) as sess:
         image_np = cv2.imread("D:/Study/dl/Pedestrian_Detection/project/test_images/3600.jpg")
         # image_np = cv2.imread("D:/images/pedestrain.png")
         cv2.imshow("input", image_np)
         print(image_np.shape)
         # image_np == [1, None, None, 3]
         image_np_expanded = np.expand_dims(image_np, axis=0)
         image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
         boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
         scores = detection_graph.get_tensor_by_name('detection_scores:0')
         classes = detection_graph.get_tensor_by_name('detection_classes:0')
         num_detections = detection_graph.get_tensor_by_name('num_detections:0')
         # Actual detection.
         (boxes, scores, classes, num_detections) = sess.run(
             [boxes, scores, classes, num_detections],
             feed_dict={image_tensor: image_np_expanded})
         # Visualization of the results of a detection.
         vis_util.visualize_boxes_and_labels_on_image_array(
               image_np,
               np.squeeze(boxes),
               np.squeeze(classes).astype(np.int32),
               np.squeeze(scores),
               category_index,
               use_normalized_coordinates=True,
               min_score_thresh=0.5,
               line_thickness=1)
         cv2.imshow('object detection', image_np)
         cv2.imwrite("D:/run_result.png", image_np)
         cv2.waitKey(0)
         cv2.destroyAllWindows()

测试效果：