大量图像的张量流预测很慢

Question

我正在关注 Tensorflow 中的图像分类教程：http://cv-tricks.com/tensorflow-tutorial/training-convolutional-neural-network-for-image-classification/。对单个图像的训练和测试工作正常。然而，我对大量图像进行预测的代码非常慢，并且消耗了 100% 的 CPU 并且几乎耗尽了内存！对于 2700 张图像，需要超过 24 小时！这是不切实际的。有没有办法像我们进行批量训练一样进行批量测试？请注意，我还需要对图像进行规范化。这是我的代码：

import tensorflow as tf
import numpy as np
import os,glob,cv2
import sys,argparse


# First, pass the path of the image
os.chdir("/somepath")

i = 0
files = glob.glob('*.jpg')
files.extend(glob.glob('*.JPG'))
totalNumber = len(files)
print("total number of images is:", totalNumber)

image_size=128
num_channels=3
text_file = open("Results.txt", "w")

for file in files:
    images = []
    filename = file
    print(filename)
    text_file.write("\n")
    text_file.write(filename)
    # Reading the image using OpenCV
    image = cv2.imread(filename)
    # Resizing the image to our desired size and preprocessing will be done exactly as done during training
    image = cv2.resize(image, (image_size, image_size),0,0, cv2.INTER_LINEAR)
    images.append(image)
    images = np.array(images, dtype=np.uint8)
    images = images.astype('float32')
    images = np.multiply(images, 1.0/255.0) 
    #The input to the network is of shape [None image_size image_size num_channels]. Hence we reshape.
    x_batch = images.reshape(1, image_size,image_size,num_channels)

    ## Let us restore the saved model 
    sess = tf.Session()
    # Step-1: Recreate the network graph. At this step only graph is created.
    saver = tf.train.import_meta_graph('pathtomymeta/my_model-9909.meta')
    # Step-2: Now let's load the weights saved using the restore method.
    saver.restore(sess, tf.train.latest_checkpoint('pathtomycheckpoints/checkpoints/'))

    # Accessing the default graph which we have restored
    graph = tf.get_default_graph()

    # Now, let's get hold of the op that we can be processed to get the output.
    # In the original network y_pred is the tensor that is the prediction of the network
    y_pred = graph.get_tensor_by_name("y_pred:0")

    ## Let's feed the images to the input placeholders
    x = graph.get_tensor_by_name("x:0") 
    y_true = graph.get_tensor_by_name("y_true:0") 
    y_test_images = np.zeros((1, 3)) #np.zeros((1, 2)) 


    ### Creating the feed_dict that is required to be fed to calculate y_pred 
    feed_dict_testing = {x: x_batch, y_true: y_test_images}
    result = sess.run(y_pred, feed_dict=feed_dict_testing)
    # result is of this format [probabiliy_of_rose probability_of_sunflower]
    print(result)
    text_file.write("\n")
    text_file.write('%s' % result[i,0])
    text_file.write("\t")
    text_file.write('%s' % result[i,1])
    text_file.write("\t")
    text_file.write('%s' % result[i,2])

text_file.close()

Answer 1

我认为您应该在代码中考虑一个非常明显的“优化”。您正在执行一个 for 循环，并且在每次迭代中，您都在加载图像，同时还加载模型、构建图形，然后进行预测。

但加载模型和构建图形实际上并不依赖于 for 循环或其中的任何变量（如输入图像）。您的 for 循环中的大部分时间可能都花在加载模型上，而不是进行实际预测。您可以使用分析器来找出答案。

所以我建议您在 for 循环之前加载模型并构建一次图形，然后在 for 循环中使用以下两行：

feed_dict_testing = {x: x_batch, y_true: y_test_images}
result = sess.run(y_pred, feed_dict=feed_dict_testing)

它应该会快很多。它可能仍然很慢，但那是因为在 CPU 上评估大型神经网络本身很慢。