TensorFlow - 使用谷歌的MNIST CNN示例，但测试自己的图像？

Question

我正在尝试使用Google的TensorFlow GitHub中的这个示例，后者构建CNN，然后对MNIST数据集进行训练和测试：

https://github.com/tensorflow/tensorflow/blob/r1.4/tensorflow/examples/tutorials/mnist/mnist_deep.py

但不是像上面那样测试MNIST的例子：

print('test accuracy %g' % accuracy.eval(feed_dict={x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0}))

我想一次一个地测试我自己的图像。我画了10张图片，Paint中的数字0 - 9，这里有几个例子：

根据此帖子进行一些更改，该帖子基于softmax Google示例：

Tensorflow - Testing a mnist neural net with my own images

这是我到目前为止所做的：

https://github.com/CDahmsTemp/MNIST_CNN_test_on_own_images/blob/master/mnist_cnn_test_on_own_images.py

我似乎无法弄清楚如何更改上面的行以使用单个.png或.jpg而不是预先配置的MNIST数据集中的多个图像。这是我到目前为止我的整个main（）函数：

#######################################################################################################################
def main():
    # import data
    print("obtaining MNIST data . . .")
    mnistData = mnistDataLib.read_data_sets("C:\\mnist", one_hot=True)

    print("building graph . . .")
    # create the model
    x = tf.placeholder(tf.float32, [None, 784])

    # define loss and optimizer
    y_ = tf.placeholder(tf.float32, [None, 10])

    # build the graph for the deep net
    y_conv, keep_prob = deepnn(x)

    with tf.name_scope('loss'):
        cross_entropy = tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv)
    # end with

    cross_entropy = tf.reduce_mean(cross_entropy)

    with tf.name_scope('adam_optimizer'):
        train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
    # end with

    with tf.name_scope('accuracy'):
        correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))
        correct_prediction = tf.cast(correct_prediction, tf.float32)
    # end with
    accuracy = tf.reduce_mean(correct_prediction)

    print("training . . .")
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        # ToDo: reduced training size for the moment to save time, restore to 20000 when get program working
        #for i in range(20000):
        for i in range(150):
            batch = mnistData.train.next_batch(50)
            if i % 100 == 0:
                train_accuracy = accuracy.eval(feed_dict={x: batch[0], y_: batch[1], keep_prob: 1.0})
                print("step " + str(i) + ", training accuracy = " + str(train_accuracy))
            # end if
            train_step.run(feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})
        # end for

        print("testing . . .")

        # this is the original code by Google to test on the MNIST data:
        # testAccuracy = accuracy.eval(feed_dict={x: mnistData.test.images, y_: mnistData.test.labels, keep_prob: 1.0})
        # print("test accuracy = " + str(testAccuracy))

        image = cv2.imread("3.jpg", cv2.IMREAD_GRAYSCALE)
        image = cv2.resize(image, (IMAGE_WIDTH, IMAGE_HEIGHT))
        imgFlattened = np.vectorize(lambda x: 255 - x)(np.ndarray.flatten(image))
        npaResult = sess.run(tf.argmax(y_, 1), feed_dict={x: [imgFlattened]})
        strResult = str(' '.join(map(str, npaResult)))
        print("result = " + strResult)
    # end with

    # write the graph to file so we can view with TensorBoard
    print("writing graph to file . . .")
    fileWriter = tf.summary.FileWriter(os.getcwd())
    fileWriter.add_graph(sess.graph)
    fileWriter.close()

    print("done !!")
# end main

其他函数调用与上面链接的Google CNN示例相同，我不想发布整个代码以防止此帖更长，如果看到整个程序是相关的，请参阅上面的GitHub链接。< / p>

目前我遇到了很多错误，但我认为最相关的是这一行：

npaResult = sess.run(tf.argmax(y_, 1), feed_dict={x: [imgFlattened]})

我收到了这个错误：

tensorflow.python.framework.errors_impl.InvalidArgumentError: You must feed a value for placeholder tensor 'Placeholder_1' with dtype float and shape [?,10]
     [[Node: Placeholder_1 = Placeholder[dtype=DT_FLOAT, shape=[?,10], _device="/job:localhost/replica:0/task:0/cpu:0"]()]]

似乎我需要向feed_dict =添加更多参数，但我不确定要添加什么。

如果我将sess.run（）行更改为：

npaResult = sess.run(tf.argmax(y_, 1), feed_dict={x: [imgFlattened], y_: mnistData.test.labels, keep_prob: 1.0})

至少程序没有崩溃，但我得到了输出：

result = 7 2 1 0 4 1 4 9 5 9 0 6 9 0 1 5 9 7 3 4 9 6 6 5 4 0 7 4 0 1 3 1 3 4 7 2 7 1 2 1 1 7 4 2 3 5 1 (many more numbers omitted)

结果行应该说

result = 3

我认为除了这一系列之外，我的一切都是正确的：

    image = cv2.imread("3.jpg", cv2.IMREAD_GRAYSCALE)
    image = cv2.resize(image, (IMAGE_WIDTH, IMAGE_HEIGHT))
    imgFlattened = np.vectorize(lambda x: 255 - x)(np.ndarray.flatten(image))
    npaResult = sess.run(tf.argmax(y_, 1), feed_dict={x: [imgFlattened]})
    strResult = str(' '.join(map(str, npaResult)))
    print("result = " + strResult)

如何更改此选项才能成功测试一个.jpg图像？