Missing attribute (storageOrder) error when compiling a simple Tensorflow model

idata · ‎11-26-2017

Hello,

I am trying to build a simple Tensorflow model and compile it for Movidius NCS, but I am running into errors and documentation is not helpful.

Code producing the model files:

#!/usr/bin/python3

import tensorflow as tf

if __name__ == '__main__':
    input_data = tf.placeholder(tf.float32, (10, 10), name='input')
    two = tf.get_variable('two', initializer=2.0)
    output_data = tf.multiply(2.0, input_data, name='output')

    saver  = tf.train.Saver()

    with tf.Session() as session:
        session.run(tf.global_variables_initializer())

        saver.save(session, '/tmp/test.tf')

After running it, these files are produced:

$ ls /tmp/test.tf*
/tmp/test.tf.data-00000-of-00001  /tmp/test.tf.index  /tmp/test.tf.meta

Compilation with errors:

$ mvNCCompile /tmp/test.tf.meta -o /tmp/test.tf.movidius
mvNCCompile v02.00, Copyright @ Movidius Ltd 2016

/usr/local/bin/ncsdk/Controllers/TensorFlowParser.py:149: DeprecationWarning: invalid escape sequence \d
  return re.sub(':\d+', '', word)
/home/jirka/.local/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow_internal.py:15: DeprecationWarning: the imp module is deprecated in favour of importlib; see the module's documentation for alternative uses
  import imp
/home/jirka/.local/lib/python3.6/site-packages/tensorflow/python/util/tf_inspect.py:45: DeprecationWarning: inspect.getargspec() is deprecated, use inspect.signature() or inspect.getfullargspec()
  if d.decorator_argspec is not None), _inspect.getargspec(target))
Traceback (most recent call last):
  File "/usr/local/bin/mvNCCompile", line 118, in <module>
    create_graph(args.network, args.inputnode, args.outputnode, args.outfile, args.nshaves, args.inputsize, args.weights)
  File "/usr/local/bin/mvNCCompile", line 108, in create_graph
    net.optimize()
  File "/usr/local/bin/ncsdk/Models/Network.py", line 197, in optimize
    self.convert_network_input_to_yxz()
  File "/usr/local/bin/ncsdk/Models/Network.py", line 229, in convert_network_input_to_yxz
    if self.storageOrder.value == StorageOrder.orderYXZ.value:
AttributeError: 'Network' object has no attribute 'storageOrder'

Changing shape of input_data to (1, 10, 10) or (1, 1, 10, 10), (1, 10, 10, 1) or (10, 10, 1) does not make any difference.

Tensorflow: 1.3.0 (I tried with 1.4.0 too. It gives more deprecation warnings, but does not affect the NCSDK error.)

NCSDK: 1.10.01.00 (According to /opt/movidius/version.txt. It does not match what mvNCCompile prints. Freshly installed.)

Any advice how to deal with it or fixes in NCSDK, if the problem is there, welcome. Thanks!

Jirka

idata · ‎11-27-2017

@jirka Thanks for bringing this to our attention. We are looking into this issue at the moment and I will let you know as soon as we find the root cause. Thank you for your patience.

idata · ‎04-07-2018

Hi,

is there any fix or workaround for this issue ?

Thanks !