Question

我对机器学习和对象检测非常陌生。我使用6000张图像训练了一个模型（我正在尝试从CCTV视频中检测事故）。首先，我尝试了SSD Mobilenet，现在该模型速度很快，但是非常不准确。现在，我使用的是faster_rcnn_resnet50_coco，现在我的准确性有所提高，但问题是它现在非常慢。（我不是在谈论训练缓慢，而是在谈论模型的实际用法是缓慢的。）

SSD Mobilenet的演示：https://imgur.com/gFUOYlx

Resnet 50的演示：https://imgur.com/we8K3Ae

我的规格：酷睿i7 6500HQ

Nvidia GTX 950M 4GB

16场公羊

我的代码（我只是从tensorflow / models / object_detection修改了object_detection.ipnyb并没有做任何特别的事情）：将numpy导入为np 导入操作系统导入six.moves.urllib作为urllib 导入系统导入tarfile 将tensorflow作为tf导入导入zipfile

from collections import defaultdict
from io import StringIO
from matplotlib import pyplot as plt
from PIL import Image
from utils import label_map_util

from utils import visualization_utils as vis_util
import cv2
def AccidentDetector(videofile):

    cap = cv2.VideoCapture(videofile)
    #MODEL_NAME = 'Accident_Detection25487-resnet'
    #MODEL_NAME = 'Accident_Detection42214-resnet'
    MODEL_NAME = 'Accident_Detection200000'
    MODEL_FILE = MODEL_NAME + '.tar.gz'
    DOWNLOAD_BASE = 'http://download.tensorflow.org/models/object_detection/'
    PATH_TO_CKPT = MODEL_NAME + '/frozen_inference_graph.pb'
    PATH_TO_LABELS = os.path.join('data', 'object-detection.pbtxt')
    NUM_CLASSES = 2
    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)
    PATH_TO_TEST_IMAGES_DIR = 'test_images'
    TEST_IMAGE_PATHS = [ os.path.join(PATH_TO_TEST_IMAGES_DIR, 'image{}.jpg'.format(i)) for i in range(1, 3) ]
    IMAGE_SIZE = (12, 8)


    count=0
    with detection_graph.as_default():
      with tf.Session(graph=detection_graph) as sess:
        while True:
          ret, image_np = cap.read()
          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')

          (boxes, scores, classes, num_detections) = sess.run(
              [boxes, scores, classes, num_detections],
              feed_dict={image_tensor: image_np_expanded})
          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,
              line_thickness=8)
          for index,value in enumerate(classes[0]):
            if scores[0,index] > 0.5:
              list1 = [[category_index.get(value)]]
              for i in list1:
                for j in i:
                  if j['name'] == 'accident':
                    '''name = "detections/frame%d.jpg"%count
                    cv2.imwrite(name, image_np)'''
                    count = count + 1

          cv2.imshow(videofile[0], cv2.resize(image_np, (1280,720)))
          if cv2.waitKey(20) & 0xFF == ord('n'):
              current = int(cap.get(cv2.CAP_PROP_POS_FRAMES))
              print(current)
              cap.set(cv2.CAP_PROP_POS_FRAMES,current+50)

          if cv2.waitKey(20) & 0xFF == ord('p'):
              current = int(cap.get(cv2.CAP_PROP_POS_FRAMES))
              print(current)
              cap.set(cv2.CAP_PROP_POS_FRAMES,current-50)

          if cv2.waitKey(20) & 0xFF == ord('q'):
            cv2.destroyAllWindows()
            break

改进张量流训练模型

0 个答案: