使用張量流中的自訓練模型標記圖像

Question

我設法在Tensorflow中用下面的圖訓練自己的模型：

在Python中，它看起來像：

with tf.name_scope("Reshaping_data") as scope:
    x = tf.reshape(x, shape=[-1, imgSize, imgSize, 1], name="inp") #(?, 48, 48, 1)

with tf.name_scope("Conv1") as scope:
    conv1 = conv2d(x, weights['wc1'], biases['bc1']) #(?, 48, 48, 32)
    conv1 = maxpool2d(conv1, k=2) #(?, 24, 24, 32)

...（更多卷積並完全連接）...

out = tf.add(tf.matmul(fc1, weights['out']), biases['out'], name="out") #(?, 43)

我使用GTSRB數據集對其進行了訓練，並保存了模型。 現在，我想用此模型標記一個新圖像。 我當前的label.py：

import tensorflow as tf
checkpoint_file = tf.train.latest_checkpoint("saved_models")
graph = tf.Graph()
with graph.as_default():
     sess = tf.Session()
     with sess.as_default():
         saver = tf.train.import_meta_graph("{}.meta".format(checkpoint_file))
         saver.restore(sess,checkpoint_file)
         inp = graph.get_operation_by_name("Reshaping_data/inp").outputs[0]
         prediction=graph.get_operation_by_name("out").outputs[0]
         input_img = tf.image.decode_jpeg(tf.read_file("/home/DB/GTSRB/Test/00021/07406.jpg"), channels=3)
         reshaped_image = tf.image.resize_image_with_crop_or_pad(tf.cast(input_img, tf.float32), 48, 48)
         float_image = tf.image.per_image_standardization(reshaped_image)
         images = tf.expand_dims(float_image, 0)
         print(sess.run(prediction,feed_dict={inp:images}))

但是在讀取feed_dict時失敗。 我究竟做錯了什么？

Traceback (most recent call last):
  File "label.py", line 23, in <module>
    print(sess.run(prediction,feed_dict={inp:images}))
  File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/client/session.py", line 767, in run
run_metadata_ptr)
  File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/client/session.py", line 925, in _run
    raise TypeError('The value of a feed cannot be a tf.Tensor object. '
 TypeError: The value of a feed cannot be a tf.Tensor object. Acceptable feed values include Python scalars, strings, lists, or numpy ndarrays.

非常感謝你！

Answer 1

Tensorflow引發錯誤，因為將Tensor / op傳遞到feed_dict 。 如果您print images您會發現看不到一個numpy數組，而是一個張量，通常在會話運行時才計算這些張量。 您需要知道傳遞給feed_dict ，例如錯誤所提到的“ Python標量，字符串，列表或numpy ndarrays”，在您的情況下將是numpy ndarray。

與其使用tensorflow讀取圖像並對其重塑， imread嘗試使用scipy，matplotlib或opencv中的imread函數，然后使用numpy進行重塑。

Answer 2

像這樣解決它：

checkpoint_file = tf.train.latest_checkpoint("saved_models")
imgSize = 48

graph = tf.Graph()
with graph.as_default():
     sess = tf.Session()
     with sess.as_default():
         saver = tf.train.import_meta_graph("{}.meta".format(checkpoint_file))
         saver.restore(sess,checkpoint_file)
         inp = graph.get_operation_by_name("Reshaping_data/inp").outputs[0]
         prediction=graph.get_operation_by_name("out").outputs[0]

         img = imread(imagepath, flatten=True)
         img = imresize(img, [imgSize, imgSize])
         img = img.astype('float32')

         img_mean = np.mean(img)
         img_var = np.std(img)
         img = (img - img_mean)/img_var

         #img = (48, 48)
         img = np.expand_dims(img, axis=2)
         #img = (48, 48, 1)
         img = np.expand_dims(img, axis=0)
         #img = (1, 48, 48, 1)

         #inp expects (?, 48, 48, 1)
         res = sess.run(prediction,feed_dict={inp:img})
         print(res)
         print(res.argmax(axis=1))