[Bug report] Exception when loading model into CPU plugin

[Lundgaard__Thomas]

Hello,

I am experiencing an issue when trying to load the attached model into open vino using the CPU plugin. I get this exception:

Node Splice14 contains empty parent edge for index 1
..\src\mkldnn_plugin\mkldnn_node.cpp:345

I strongly suspect this is caused by a Concat node that takes the same input on the to input ports. Here is a code snippet to reproduce:

auto plugin = PluginDispatcher({ L"" }).getPluginByDevice("CPU");
CNNNetReader networkReader;
networkReader.ReadNetwork("saved.xml");
networkReader.ReadWeights("saved.bin");
auto cnnNetwork = networkReader.getNetwork();
cnnNetwork.setBatchSize(1);
auto executableNetwork = plugin.LoadNetwork(cnnNetwork, {});

Best Regards,

Thomas

[nikos1]

Hi Thomas,

Attached IR seems to load fine on Linux InferenceEngine API version 1.4 Build 19154 SDK computer_vision_sdk_2018.5.445 . Could it be an environment issue - can you see all libraries and plug-ins? What OS and SDK version?

nikos

[Lundgaard__Thomas]

Hi Nikos,

Thanks for testing this out on your machine.

I am on Windows 10 with computer vision sdk version 2018.5.456. I have also just tried with 2018.5.445, and get the same issue.

I have only installed the CPU and GPU plugin (although my computer doesn't have an intel integrated GPU).

What do you mean by "can you see all libraries and plug-ins"?

I am able to run inference with other models.

Thomas

[Shubha_R_Intel]

Hi Thomas. I'm with Nikos. I tried this on Windows,  and I could not reproduce this issue. Let me tell you what I did...I used the classification sample and passed in your model (from the zip file). Rather than hard-coding stuff as you did in your sample, I pass them in via command-line arguments as the classification sample does. Works perfectly fine.

Did you make sure to execute bin\setupvars.bat ?

[Lundgaard__Thomas]

Hi,

I just tried using the classification sample, and I still get the same exception. Note that the vanilla classification sample exits prior to loading the model into the plugin (throws "Incorrect output dimensions for classification model"). So I had to remove that check before I could expose the behaviour using the classification sample.

I also made sure to executed setupvars.bat before running the program, but it made no difference.

- Thomas

[Shubha_R_Intel]

I was able to reproduce this. I will file a bug Thomas. Thank you !