IR model does not work in Myriad-2

[Katakis__Sofoklis]

Hello,

I have trained an model with pytorch toolkit of openvino_training_extensions. Afterwards I successfully converted it to ONNX with the nncf utility existing in the abovementioned toolkit.

Finally I converted the model in the IR representation  (FP16 and FP32 format) with the openvino model optimizer utilities.

The problem that I have encountered is the following.

I tested both of these formats with the benchmark app in my Ubuntu CPU and they are working. When later I try to test them in the MYRIAD-2 NCS stick then an error breaks the benchmark routine.

The error (which I have not found it in the forums) is the following:

[ ERROR ] AssertionFailed: data->numConsumers() > 0

I attach you the models with other related informations.

Thank you in advance for your time.

[Shubha_R_Intel]

Dear Katakis, Sofoklis,

Did you try on OpenVino 2019R3 ? It was just released this week. If not please do so, and report back on this forum.

Thanks !

Shubha

[Katakis__Sofoklis]

Dear Shubha,

Thank you for your answer.

Yes I tried. The same thing.

[Shubha_R_Intel]

Dear Katakis, Sofoklis,

Thanks for trying 2019R3 anyway. Let me debug this and get back to you.

Thanks for your patience !

Shubha

[Sahira_Intel]

Hi Sofoklis,

If you still require assistance on this issue, can you please provide the commands you used to convert your model, and also the full command you used to deploy your model. I see you also provided a smaller file on your Github ticket here. 

Sincerely,

Sahira 

[Katakis__Sofoklis]

Dear Sahira,

The command that I used to deploy the model is in log files. I copy it from there.

 "./intel64/Debug/benchmark_app  -i ~/test_im -m /path_to_the_model/model_FP16.xml  -d MYRIAD -b 1"

The command that I used to transform the model is via onnx. I tried various combinations but non was working. An example is the following.

python mo_onnx.py -m Onnx_model_path -n Model_name -o Output_path  --data_type FP16 --scale 255 --input_shape [1,3,300,300]

Hope that I helped,

K. Sofoklis

[Sahira_Intel]

Hi,

It looks like there was a bug in the VPU plugin. This issue will be fixed in the next OpenVINO release. 

Please let me know if you have any further questions!

Best Regards,

Sahira

[赵__明]

Hello！I have a very similar problem to this,my IR model work with CPU but not MYRIAD,and the error message is(using  2019R3 ):

  File "classification_sample_new.py", line 349, in main
    exec_net = ie.load_network(network=net, device_name=args.device)
  File "ie_api.pyx", line 85, in openvino.inference_engine.ie_api.IECore.load_network
  File "ie_api.pyx", line 92, in openvino.inference_engine.ie_api.IECore.load_network
RuntimeError: AssertionFailed: _allocatedIntermData.count(topParent) > 0

For more details please see:

https://software.intel.com/zh-cn/node/832716

Thanks for attention!

[Sahira_Intel]

Hi, 

This error is caused by a bug in the VPU plugin that will be resolved in the next OpenVINO release.

Best Regards,

Sahira 

[Thierry__Lapierre]

Hello, I have the same problem with a Myriad V1. My Network works in CPU mode and when I switch to 'MYRIAD' mode I get this error message.

You can find in attach files the optimise Network and the Python file (inferenceV2myriad.py)

I work with openvino 2019.3.334

Can you say when the next release is Ok for correct this bug ?

Thank's

error message :

python3 inferenceV2myriad.py 
Input Layer:  im0
Output Layer:  2848/Split.1
Input Shape:  [1, 3, 256, 512]
Output Shape:  [1, 6, 64, 128]
Available Devices:  ['MYRIAD']
Traceback (most recent call last):
  File "inferenceV2myriad.py", line 65, in <module>
    OpenVinoExecutable = OpenVinoIE.load_network(network=OpenVinoNetwork,device_name='MYRIAD')
  File "ie_api.pyx", line 85, in openvino.inference_engine.ie_api.IECore.load_network
  File "ie_api.pyx", line 92, in openvino.inference_engine.ie_api.IECore.load_network
RuntimeError: AssertionFailed: _allocatedIntermData.count(topParent) > 0