Intel® Distribution of OpenVINO™ Toolkit
Community support and discussions about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all things computer vision-related on Intel® platforms.
5532 Discussions

Network with Matmul cannot be used on Myriad

Jurado__Michael
Beginner
372 Views

 

I have a custom model that runs perfectly fine on CPU but fails with Myriad with this error:

 

src/file.py:68: DeprecationWarning: Reading network using constructor is deprecated. Please, use IECore.read_network() method instead
  net = IENetwork(model=model_xml, weights=model_bin)
src/file.py:69: DeprecationWarning: IEPlugin class is deprecated. Please use IECore class instead.
  plugin = IEPlugin(device=device)
Traceback (most recent call last):
  File "src/file.py", line 74, in <module>
    exec_net = plugin.load(network=net, num_requests = 2)
  File "ie_api.pyx", line 1352, in openvino.inference_engine.ie_api.IEPlugin.load
  File "ie_api.pyx", line 1362, in openvino.inference_engine.ie_api.IEPlugin.load
RuntimeError: Failed to reshape Network: Check 'arg0_shape[axis_index_arg0].compatible(arg1_shape[axis_index_arg1])' failed at C:..dot.cpp:130:
While validating node 'v0::Dot Dot_3170(VariadicSplit_2866[0]:f32{1,128}, Relu_2873[0]:f32{1,50}) -> (dynamic?)':
Paired axes (axis 1 from arg0, axis 0 from arg1) do not have same length (arg0 shape: {1,128}, arg1 shape: {1,50}, reduction axes count: 1).

 

 

Additional Information:

Tested on openvino_2020.3.194, windows 10

-I also tried plugin.set_config({'VPU_HW_STAGES_OPTIMIZATION': 'NO'}) but no dice. 

 

0 Kudos
2 Replies
JAVIERJOSE_A_Intel
372 Views

Hi Michael,

 

Thanks for reaching out.

Could you please answer the following questions:

  • Which OS are you using?
  • Can you share with us the files needed to test it from our end (model, input, code sample)? If you do not want to share it publicly, you can share it via private message.

 

Regards,

Javier A.

Jurado__Michael
Beginner
372 Views

I am using WIndows 10. I'l get back to you shortly about a code sample.

Reply