[Error 17] Toolkit Error

[idata]

I want to compile my caffe model based on CNN, however I got following error, I can't keep going more.

If you have any solution or you know why mvNCCompile give me following error message, please help me.my environment

• host: ubuntu16.04
• NCSDK V2
• mvNCCompile version: mvNCCompile v02.00, Copyright @ Intel Corporation 2017
• NCSDK example(AlexNet and GoogleNet) compile and run: SUCCESS

error message

mvNCCompile deploy.prototxt -w model.caffemodel -s 12

/usr/local/bin/ncsdk/Controllers/Parsers/TensorFlowParser/Convolution.py:46: SyntaxWarning: assertion is always true, perhaps remove parentheses?

assert(False, "Layer type not supported by Convolution: " + obj.type)/usr/local/bin/ncsdk/Controllers/Parsers/Phases.py:322: SyntaxWarning: assertion is always true, perhaps remove parentheses? assert(len(pred) == 1, "Slice not supported to have >1 predecessors")mvNCCompile v02.00, Copyright @ Intel Corporation 2017

Eltwise Eltwise1 does not have coefficient. Use the default one

Eltwise Eltwise2 does not have coefficient. Use the default oneEltwise Eltwise3 does not have coefficient. Use the default oneEltwise Eltwise4 does not have coefficient. Use the default oneEltwise Eltwise5 does not have coefficient. Use the default oneEltwise Eltwise6 does not have coefficient. Use the default oneEltwise Eltwise7 does not have coefficient. Use the default oneEltwise Eltwise8 does not have coefficient. Use the default oneEltwise Eltwise9 does not have coefficient. Use the default oneFusing Pad and Convolution2DFusing BatchNorm and Scale after ConvolutionReplacing BN with Bias&ScaleFusing Permute and FlattenFusing Eltwise and Relu[Error 17] Toolkit Error: Internal Error: Could not build graph. Missing link: Fusing into a layer with multiple successors is not supported: (Output#140)

[idata]

@jongseong.kim Can you provide a link to your model? I'd like to take a closer look at the network. Thanks.

[idata]

Hi, Tome

Thanks for your support and I uploaded the model and prototxt files in following URL.

• Model file: https://drive.google.com/open?id=1ZB4XQw5e6t-40PKMwQ65ANOuQsHRlEN2
• Prototxt file: https://drive.google.com/open?id=1ased3lpvJbTvHKpNHAOKu0trJMWNrwqQ

[idata]

@jongseong.kim I was unable to compile the model. I received an error: [Error 17] Toolkit Error: Internal Error: Could not build graph. Missing link: ReLU2. Please double check your model and send it again please. Thanks.

Here is the log for my run:

$ mvNCCompile deploy.prototxt -w model.caffemodel -s 12 mvNCCompile v02.00, Copyright @ Movidius Ltd 2016 [Error 17] Toolkit Error: Internal Error: Could not build graph. Missing link: ReLU2

[idata]

@Tome_at_Intel

I think that mvNCCompile version is different with our case.In case of ours, it shows like that "mvNCCompile v02.00, Copyright @ Intel Corporation 2017".

And I downloaded the files which I uploaded, I run again via "mvNCCompile deploy.prototxt -w model.caffemodel -s 12".

It shows the following log:

====================== LOG start ======================

mvNCCompile deploy.prototxt -w model.caffemodel -s 12/usr/local/bin/ncsdk/Controllers/Parsers/TensorFlowParser/Convolution.py:46: SyntaxWarning: assertion is always true, perhaps remove parentheses? assert(False, "Layer type not supported by Convolution: " + obj.type)/usr/local/bin/ncsdk/Controllers/Parsers/Phases.py:322: SyntaxWarning: assertion is always true, perhaps remove parentheses? assert(len(pred) == 1, "Slice not supported to have >1 predecessors")mvNCCompile v02.00, Copyright @ Intel Corporation 2017

Eltwise Eltwise1 does not have coefficient. Use the default one

Eltwise Eltwise2 does not have coefficient. Use the default oneEltwise Eltwise3 does not have coefficient. Use the default oneEltwise Eltwise4 does not have coefficient. Use the default oneEltwise Eltwise5 does not have coefficient. Use the default oneEltwise Eltwise6 does not have coefficient. Use the default oneEltwise Eltwise7 does not have coefficient. Use the default oneEltwise Eltwise8 does not have coefficient. Use the default oneEltwise Eltwise9 does not have coefficient. Use the default oneFusing Pad and Convolution2DFusing BatchNorm and Scale after ConvolutionReplacing BN with Bias&ScaleFusing Permute and FlattenFusing Eltwise and Relu[Error 17] Toolkit Error: Internal Error: Could not build graph. Missing link: Fusing into a layer with multiple successors is not supported: (Output#140)====================== LOG end ======================

Please check again.

Thanks so much for your support.

[idata]

@jongseong.kim There seems to be multiple issues with running this network. I'm getting wrong results right after the first concat. Will look into this some more.

[idata]

@Tome_at_Intel

Thanks for your reply, and I will wait for your answer to use Movidius that is a very nice.

[idata]

@jongseong.kim It seems like NCSDK is having issues with the Eltwise and ReLU layers. Eltwise1 layer seems to have multiple outputs and the NCSDK is having issues processing this with the ReLU5 layer. Looks like a bug on our end. Thanks for reporting this.

[idata]

Is there any resolve?

I got the exact same problem here.

[idata]

Hi @DableUTeeF

Have you tried updating to the latest NCSDK 2.10.01?

Best Regards,

Sahira

[idata]

@Sahira_at_Intel

I compiled it with version 2 like in instruction but how could I check what exactly version it is?I just clone the repo and compiled it last month though.

[idata]

It's very interesting to see that even this page https://ncsforum.movidius.com/discussion/370/intel-ncs-troubleshooting-help-and-guidelines doesn't have how to check the installed software version.

[idata]

Hi @DableUTeeF

You can navigate to the directory that you initally installed the NCSDK in and see what version you have installed from there!

Best Regards,

Sahira

[idata]

Hi @Sahira_at_Intel

It shows only ncsdk2 for git branch in the repo I use, the latest possible in this repo is v2.10.01.01, I think it's already v2.10.There's nothing about version in /usr/local/bin/bcsdk at all.