Implementation vs real world example

[engahmed1190]

Hello,

I would like to know the importance of having annotation images in the testing of the model , why we don't provide RBG images only

Basically this is the implementation

valid_images, valid_annotations = validation_dataset_reader.get_consecutive_batch(FLAGS.batch_size)
pred = sess.run(pred_annotation, feed_dict={image: valid_images, annotation: valid_annotations,
                                                    keep_probability: 1.0})

why don't we only do that

pred = sess.run(pred_annotation, feed_dict={image: valid_images, keep_probability: 1.0})

@txzhao can you help on this please

[txzhao]

Hi,

Sorry for the late reply. I'm kind of occupied by some other projects and don't have much time to check the code. From what I've seen so far, I think you are right. We could simplify this line of code. I must have been too lazy or hurry to think about it clearly when I coded this part. But I think it won't affect the testing results.

[engahmed1190]

Hello Tianxiao

i have tried to load the model using this code

from __future__ import print_function
import tensorflow as tf
import numpy as np


import inception_v3_fcn

IMAGE_SIZE = 224

NUM_OF_CLASSESS = 22


def load_image(filename):
  with open(filename) as f:
    return np.array(f.read())|

sess = tf.Session()

print("Setting up Saver...")

saver = tf.train.import_meta_graph('./logs/allmodel.ckpt-100000.meta')
print("meta graph loaded")

saver.restore(sess,tf.train.latest_checkpoint('./logs'))

all_vars = tf.get_collection('vars')
'''
for v in tf.global_variables():
    print(v.name)
'''
image = tf.placeholder(tf.float32, shape=[None, IMAGE_SIZE, IMAGE_SIZE, 3], name="input_image")
tf.summary.image("input_image", image, max_outputs=2)

pred_annotation, logits, end_points = inception_v3_fcn.inception_v3_fcn(image, num_classes=NUM_OF_CLASSESS, dropout_keep_prob=1.0)

tf.summary.image("pred_annotation", tf.cast(pred_annotation, tf.uint8), max_outputs=2)
pred = sess.run(pred_annotation, feed_dict={image: load_image('actor.jpg').flatten()})

i have get an error related to

Traceback (most recent call last):
  File "loading_model.py", line 42, in <module>
    pred = sess.run(pred_annotation, feed_dict={image: load_image('actor.jpg').flatten()})
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 895, in run
    run_metadata_ptr)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 1093, in _run
    np_val = np.asarray(subfeed_val, dtype=subfeed_dtype)
  File "/usr/local/lib/python2.7/dist-packages/numpy/core/numeric.py", line 531, in asarray
    return array(a, dtype, copy=False, order=order)
ValueError: could not convert string to float: ����

i think it related to the image feed , actually i need to understand this part to feed the image with real data , can you please advise how i can achieve that

Thanks in advance
@txzhao

[txzhao]

Hi,

I don't remember I did the flatten() operation to the input image. You can try get rid of that, and if still errors show up, I would suggest checking the code inside BatchDatsetReader.py. Be careful about the shape of input images.

[engahmed1190]

@txzhao
Hello ,

i have found that you use

image = misc.imread("actor.jpg")
temp_resize_image = misc.imresize(image,[224, 224], interp='nearest')
resize_image = temp_resize_image.reshape([-1, 224, 224, 3])

but still something missing , that is the error

Traceback (most recent call last):
  File "inception_FCN.py", line 283, in <module>
    tf.app.run()
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/platform/app.py", line 48, in run
    _sys.exit(main(_sys.argv[:1] + flags_passthrough))
  File "inception_FCN.py", line 281, in main
    pred = sess.run(pred_annotation, feed_dict={image:hash(tuple(resize_image))})
TypeError: unhashable type: 'numpy.ndarray'

[txzhao]

sorry for the late reply. I don't quite understand why you want to hash your input image. Could you try getting rid of hash(tuple(...)) part?
Btw, as for the shape of image, it should have following form: [number_of_images, width_of_image, height_of_image, number_of_channels].

[engahmed1190]

@txzhao
i hope you are still available , i manged to find out the issue . name image is used somewhere else
i want to save the predicted image correctly
i am using that

pred = np.squeeze(pred, axis=3)
misc.imsave(pred,"pred.png" )

i got this error

Traceback (most recent call last):
  File "inception_FCN.py", line 287, in <module>
    tf.app.run()
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/platform/app.py", line 48, in run
    _sys.exit(main(_sys.argv[:1] + flags_passthrough))
  File "inception_FCN.py", line 284, in main
    misc.imsave("pred.png", pred)
  File "/usr/local/lib/python2.7/dist-packages/scipy/misc/pilutil.py", line 197, in imsave
    im = toimage(arr, channel_axis=2)
  File "/usr/local/lib/python2.7/dist-packages/scipy/misc/pilutil.py", line 289, in toimage
    raise ValueError("'arr' does not have a suitable array shape for "
ValueError: 'arr' does not have a suitable array shape for any mode.

[txzhao]

Hi, yes I'm available. I checked this scipy.misc.imsave function and it seems that it could only handle specific image matrices (with shape HxW, HxWx3, HxWx4). Please check if pred has the right shape, and an alternative is using matplotlib.pyplot.savefig to save your result. I used it in colorize.py for your information.

[engahmed1190]

(1, 224, 224) is the prediction shape . think its ok

[txzhao]

shouldn't the shape be (224, 224) or (224, 224, 3) instead of (1, 224, 224)?
https://docs.scipy.org/doc/scipy/reference/generated/scipy.misc.imsave.html

[engahmed1190]

i have made a dirty trick

        pred = np.squeeze(pred, axis=3)
	pred = np.squeeze(pred, axis=0)

i get this picture

i want to make a simple colorization

       plt.imshow(pred, cmap=cmap, vmin=0, vmax=22)
	plt.title('prediction')

[txzhao]

this should work. don't forget to denote cmap as cmap = plt.cm.jet.