sess.run()太慢了

Question

Tensorflow对象检测模块的sess.run()函数需要大约2.5秒的时间来检测600x600图像中的圈圈。我怎样才能加速这段代码呢？

def run(image, detection_graph):

with detection_graph.as_default():
    with tf.Session(graph=detection_graph) as sess:
        # Definite input and output Tensors for detection_graph
        image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
        # Each box represents a part of the image where a particular object was detected.
        detection_boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
        # Each score represent how level of confidence for each of the objects.
        # Score is shown on the result image, together with the class label.
        detection_scores = detection_graph.get_tensor_by_name('detection_scores:0')
        detection_classes = detection_graph.get_tensor_by_name('detection_classes:0')
        num_detections = detection_graph.get_tensor_by_name('num_detections:0')

        # the array based representation of the image will be used later in order to prepare the
        # result image with boxes and labels on it.
        image_np = image
        # Expand dimensions since the model expects images to have shape: [1, None, None, 3]
        image_np_expanded = np.expand_dims(image_np, axis=0)
        # Actual detection.
        print("2")
        start_time = datetime.datetime.now()
        (boxes, scores, classes, num) = sess.run(
          [detection_boxes, detection_scores, detection_classes, num_detections],
          feed_dict={image_tensor: image_np_expanded})
        end_time = datetime.datetime.now()
        diff = (end_time - start_time).total_seconds()*1000
        print (diff)
        print("3")

        return boxes[0], scores[0]
        #print scores
        #print classes

Answer 1

您的sess.run执行时间对于第一次运行是正常的，在此之后，它的运行速度可能会快100倍(不是开玩笑的)。

关键是重用会话，在您的示例中，我将添加另一个图像评估，并度量该时间，并检查性能是否有所改善，例如：

# all your prev code here
print (diff)
print("3")

image_np = image2 # get another image from somewhere
image_np_expanded = np.expand_dims(image_np, axis=0)
start_time = datetime.datetime.now()

(boxes, scores, classes, num) = sess.run(
          [detection_boxes, detection_scores, detection_classes, num_detections],
          feed_dict={image_tensor: image_np_expanded})
end_time = datetime.datetime.now()

diff = (end_time - start_time).total_seconds()*1000
print("Detection #2")
print(diff)

因此，您不需要GPU或较小的图像(目前)，只需要“热身”的会话，并使用它的所有预测。

我目前在一个测试环境中做了一个非常温和的设置，Ubuntu的最后一个版本运行在VirtualBox上，没有GPU (MobileNet2 + COCO )，一旦会话“温暖”，我得到的时间是相当不错的。

--- 3.7862255573272705 seconds ---
--- 0.21631121635437012 seconds ---
--- 0.1784508228302002 seconds ---

注意第一个慢执行时间，最后一个是大小为1050*600的图像。