Depthwise Conv Not Working Correctly

[idata]

I am using Tensorflow 1.9.0 and NCSDK (v2) v2.8.1.2 in Ubuntu 16.04.

============================================================================================

I have the very simple tensorflow graph:import tensorflow as tfimport numpy as npimport sys

inputs = tf.placeholder(tf.float32, (1,402,202,60), name="input")

weights = tf.get_variable("weights", [3, 3, 60, 1], dtype=tf.float32, initializer=tf.random_uniform_initializer)outputs = tf.nn.depthwise_conv2d(inputs, weights, [1, 1, 1, 1], "VALID", name="output", data_format='NHWC')

testIn = np.ones((1,402,202,60))

saver = tf.train.Saver()with tf.Session() as sess:…………sess.run(tf.global_variables_initializer())…………tfOut = sess.run(outputs, feed_dict={inputs:testIn})…………save_path = saver.save(sess, "./testGraph/depthwise")print(tfOut.shape)

============================================================================================

The tensorflow program runs OK and created the "meta" graph in "./testGraph/depthwise.meta"

Now, when I run the following commands:

"cd testGraph && mvNCCheck depthwise.meta -s 12 -in input -on output"

I got the following output:

============================================================================================

mvNCCheck v02.00, Copyright @ Intel Corporation 2017

_* Info: No Weights provided. inferred path: depthwise.data-00000-of-00001_*

shape: (1, 402, 202, 60)res.shape: (1, 400, 200, 60)TensorFlow output shape: (400, 200, 60)Blob generatedUSB: Transferring Data…/usr/lib/python3.5/site-packages/mvnc/mvncapi.py:418: DeprecationWarning: The binary mode of fromstring is deprecated, as it behaves surprisingly on unicode inputs. Use frombuffer insteadUSB: Myriad Execution FinishedUSB: Myriad Connection Closing.USB: Myriad Connection Closed.Result: (400, 200, 60)1) 2219925 6.0942) 4589229 5.8633) 3984105 5.824) 480525 5.7735) 1113525 5.73Expected: (400, 200, 60)1) 2219925 6.09412) 4589229 5.8631873) 3984105 5.8211014) 480525 5.7733215) 1113525 5.7299337

Obtained values

Obtained Min Pixel Accuracy: 68.20873022079468% (max allowed=2%), Fail

Obtained Average Pixel Accuracy: 2.201648987829685% (max allowed=1%), Fail Obtained Percentage of wrong values: 6.6650625% (max allowed=0%), Fail Obtained Pixel-wise L2 error: 8.806281379565066% (max allowed=1%), Fail Obtained Global Sum Difference: 644019.375

============================================================================================

Noticed that, there are lots of errors in the sum of difference and most of the test have failed. I am using the 3x3 depthwise filter in this case.

Any ideas on how to address this issue in the NCS???

Thanks a lot for all the helps and supports….