Different results from Original Tensorflow model and Inference from NCS

[idata]

Hello ,

I have been getting wrong results in comparison with the original tensorflow model. I followed the guidance to create the tensorflow model without unneccesary layers (such as dropout) and the created model works perfectly when predicted with GPU. Then once I create the freeze graph, and use it to predict my dataset, i get completely different results. The things is that mvNCCheck passes every test(see below details). But the graph created with mvNCCompile produces completely wrong results. Kindly help me on this.

`sangathamilan@sangathamilan:~/Project/without_concat$ mvNCCheck ./NCS1/inference.meta -s 12 -in=input -on=softmax_out -is 100 25 -cs 0,1,2 -ec

mvNCCheck v02.00, Copyright @ Movidius Ltd 2016

/usr/local/lib/python3.5/dist-packages/tensorflow/python/framework/ops.py:939: DeprecationWarning: builtin type EagerTensor has no module attribute

EagerTensor = c_api.TFE_Py_InitEagerTensor(_EagerTensorBase)/usr/local/lib/python3.5/dist-packages/tensorflow/python/util/tf_inspect.py:55: DeprecationWarning: inspect.getargspec() is deprecated, use inspect.signature() instead if d.decorator_argspec is not None), _inspect.getargspec(target))/usr/local/lib/python3.5/dist-packages/tensorflow/python/util/tf_inspect.py:55: DeprecationWarning: inspect.getargspec() is deprecated, use inspect.signature() instead if d.decorator_argspec is not None), _inspect.getargspec(target))/usr/local/bin/ncsdk/Controllers/FileIO.py:52: UserWarning: You are using a large type. Consider reducing your data sizes for best performance "Consider reducing your data sizes for best performance\033[0m")USB: Transferring Data…USB: Myriad Execution FinishedUSB: Myriad Connection Closing.USB: Myriad Connection Closed.Result: (1, 1, 2)1) 1 0.734382) 0 0.26514Expected: (1, 2)1) 1 0.7287142) 0 0.271286 Obtained values

** Obtained Min Pixel Accuracy: 0.8437934331595898% (max allowed=2%), Pass

Obtained Average Pixel Accuracy: 0.8102884516119957% (max allowed=1%), Pass Obtained Percentage of wrong values: 0.0% (max allowed=0%), Pass Obtained Pixel-wise L2 error: 0.8109808796745067% (max allowed=1%), Pass Obtained Global Sum Difference: 0.011809378862380981**

`

[idata]

Hi @sangathamilan

What application are you using? Can you provide the code you're using? I'd like to try to reproduce this issue to see what kind of results I'm getting. Bad results sometimes mean there could be an error in the pre processing.

Best Regards,

Sahira