Regarding defining input_placeholder to predict ov

问题:

In the following function, author aims to run prediction over a downloaded image based on vgg model.

with tf.Graph().as_default():

url = ("https://upload.wikimedia.org/wikipedia/commons/d/d9/First_Student_IC_school_bus_202076.jpg")

image_string = urllib2.urlopen(url).read()
image = tf.image.decode_jpeg(image_string, channels=3)

image_float = tf.to_float(image, name='ToFloat')

# Subtract the mean pixel value from each pixel
processed_image = _mean_image_subtraction(image_float,
                                          [_R_MEAN, _G_MEAN, _B_MEAN])

input_image = tf.expand_dims(processed_image, 0)

with slim.arg_scope(vgg.vgg_arg_scope()):


    logits, _ = vgg.vgg_16(input_image,
                           num_classes=1000,
                           is_training=False,
                           spatial_squeeze=False)
pred = tf.argmax(logits, dimension=3)

init_fn = slim.assign_from_checkpoint_fn(
    os.path.join(checkpoints_dir, 'vgg_16.ckpt'),
    slim.get_model_variables('vgg_16'))

with tf.Session() as sess:
    init_fn(sess)
    segmentation, np_image, np_logits = sess.run([pred, image, logits])

I have been trying to predict over an existing image read via opencv, the only modification I made is to read image via cv2, add input_placeholder, and modify sess.run correspondingly. However, I got the following error message:

segmentation, np_image, np_logits = sess.run([pred,logits],feed_dict={input_placeholder:image})
ValueError: need more than 2 values to unpack

Would you like to let me know which modification I made is wrong?

with tf.Graph().as_default():

image = cv2.imread('/data/cat.jpg',cv2.IMREAD_UNCHANGED)
input_placeholder = tf.placeholder(tf.float32,shape = [image.shape[0],image.shape[1],image.shape[2]])
image_float = np.float32(image)

# Subtract the mean pixel value from each pixel
processed_image = _mean_image_subtraction(image_float,[_R_MEAN, _G_MEAN, _B_MEAN])

input_image = tf.expand_dims(processed_image, 0)

with slim.arg_scope(vgg.vgg_arg_scope()):

    logits, _ = vgg.vgg_16(input_image,
                           num_classes=1000,
                           is_training=False,
                           spatial_squeeze=False)


pred = tf.argmax(logits, dimension=3)

init_fn = slim.assign_from_checkpoint_fn(
    os.path.join(checkpoints_dir, 'vgg_16.ckpt'),
    slim.get_model_variables('vgg_16'))

with tf.Session() as sess:
    init_fn(sess)
    segmentation, np_image, np_logits = sess.run([pred,logits],feed_dict={input_placeholder:image})

回答1:

For reference first take a look to the official documentation: https://www.tensorflow.org/api_docs/python/tf/Session#run

For each graph element passed to the fetches parameter of sess.run() you get one value returned. In your case you are passing the following list as fetches: [pred,logits] so sess.run([pred,logits], ...) will return 2 values: the result of running the pred op and the logits op.

Citing the documentation

The value returned by run() has the same shape as the fetches argument, where the leaves are replaced by the corresponding values returned by TensorFlow.

However in this line

segmentation, np_image, np_logits = sess.run([pred,logits],feed_dict={input_placeholder:image})

you are trying to assign these 2 values to 3 different python variables (segmentation, np_image, np_logits) hence you get the ValueError.

If you look at the original example you provided, the final line is:

segmentation, np_image, np_logits = sess.run([pred, image, logits])

To mimic the original example try removing the np_image declaration from your code like so:

segmentation, np_logits = sess.run([pred,logits],feed_dict={input_placeholder:image})

回答2:

is a python error:

segmentation, np_image, np_logits = sess.run([pred,logits],feed_dict={input_placeholder:image})

should be

np_image, np_logits = sess.run([pred,logits],feed_dict={input_placeholder:image})

you are only passing two fetches to the run function