Segmentation sample crashes: FCN

[Julien1]

I successfully converted a TensorFlow Inception V1 network to IR (on ubuntu) and ran it with the classification_sample using both CPU (MKLDNN) and GPU (CLDNN) plugins, under Windows 10.

However, when I tried to run the segmentation sample with a fully connected net it crashes while trying to load the IR .xml file. I generated the IR files using:

ModelOptimizer -p FP32 -w fcn8s-heavy-pascal.caffemodel -d deploy.prototxt -i -b 1

I got the the caffemodel from here:
http://dl.caffe.berkeleyvision.org/fcn8s-heavy-pascal.caffemodel

and the deploy.prototxt from there:
https://github.com/shelhamer/fcn.berkeleyvision.org/tree/master/voc-fcn8s

ModelOptimizer generated default.xml and default.bin. The XML looks ok. Here is a link to it on my dropbox:

https://www.dropbox.com/s/yqq21mzaobt45kv/default.xml?dl=0

Running segmentation_sample on Windows 10 with:

segmentation_sample.exe -i image.bmp -m "default.xml" -d "CPU"

(or with -d "GPU") 

causes the following crash when reading the network in main.cpp, Line 142 of the sample/segmentation_sample:

network.ReadNetwork(FLAGS_m);

throws an exception

Call Stack:

  ucrtbased.dll!00007ffcd3f6765c() Unknown Non-user code. Cannot find or open the PDB file.

  ucrtbased.dll!00007ffcd3f675a7() Unknown Non-user code. Cannot find or open the PDB file.

  inference_engine.dll!00007ffcea87b809() Unknown No symbols loaded.

  inference_engine.dll!00007ffcea8acd26() Unknown No symbols loaded.

  inference_engine.dll!00007ffcea83c3f9() Unknown No symbols loaded.

  inference_engine.dll!00007ffcea82d697() Unknown No symbols loaded.

  inference_engine.dll!00007ffcea82cba4() Unknown No symbols loaded.

> segmentation_sample.exe!InferenceEngine::CNNNetReader::ReadNetwork(const std::basic_string<char,std::char_traits<char>,std::allocator<char> > & filepath) Line 59 C++ Symbols loaded.

  segmentation_sample.exe!main(int argc, char * * argv) Line 143 C++ Symbols loaded.

  [External Code] Annotated Frame

Immediate Window:

Exception thrown at 0x00007FFD0B9B3FB8 in segmentation_sample.exe: Microsoft C++ exception: InferenceEngine::details::InferenceEngineException at memory location 0x000000338EEFC9A0.
Exception thrown at 0x00007FFD0B9B3FB8 in segmentation_sample.exe: Microsoft C++ exception: InferenceEngine::details::InferenceEngineException at memory location 0x000000338EEFC9A0.
Exception thrown at 0x00007FFD0B9B3FB8 in segmentation_sample.exe: Microsoft C++ exception: InferenceEngine::details::InferenceEngineException at memory location 0x000000338EEFC9A0.
Debug Assertion Failed!

Program: C:\WINDOWS\SYSTEM32\MSVCP140D.dll
File: C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\include\vector
Line: 1234

Expression: vector subscript out of range

For information on how your program can cause an assertion
failure, see the Visual C++ documentation on asserts.

(Press Retry to debug the application)
Debug Assertion Failed!

Program: ...K\1.0\Inference Engine\bin\intel64\Debug\inference_engine.dll
File: C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\include\vector
Line: 1235

Expression: "Standard C++ Libraries Out of Range" && 0

For information on how your program can cause an assertion
failure, see the Visual C++ documentation on asserts.

(Press Retry to debug the application)

Any suggestions appreciated.

Thanks,

Julien.

[Julien1]

Out of interest I ran the same network under Ubuntu 16.04 and I was able to run the network with both GPU - clDNN and CPU - mklDNN plugins - looks like a Windows issue relating to loading the network. I am only interested in inferencing under Windows using the Intel GPU so I still need to find a way to resolve this issue. Any help appreciated !

[Anna_B_Intel]

Hi Julien, 

I generated the identical intermediate representation (.bin and .xml files) and successfully build and ran segmentation_sample on both Windows and Ubuntu. It's probably environmental issue. Which version of Computer Vision SDK do you use?

Could you please share binary of the segmentation_sample and input image with me? 

Best wishes, 

Anna

[Julien1]

Thanks Anna:

I checked and I was using the 2017 R2 SDK. I updated to 2017 R3 SDK and it no longer crashes. Thanks for pointing me in the right direction.

Julien.