face-detection-retail-0004 zero dimension error when reading on RPI4 + NCS2 (2020.3.220)

[Hawkes__Rycharde]

I assume this model should run fine but I get this error when reading it:

  File "ie_api.pyx", line 136, in openvino.inference_engine.ie_api.IECore.read_network
  File "ie_api.pyx", line 157, in openvino.inference_engine.ie_api.IECore.read_network
RuntimeError: fc7_mbox_priorbox/naked_not_unsqueezed has zero dimension that is not allowable

 Is there a different version of the model I should be using on the RPI?

[Hawkes__Rycharde]

And, annoyingly, to answer my own question you have to use the 2019 version of the model: InferenceEngineException-on-raspberrypi-Product-of-output-shape 

[Luis_at_Intel · Accepted solution]

Hi @Hawkes__Rycharde,

This is a known issue with the pre-built packages available for Raspbian* in Intel® Open Source Technology Center. In addition to the workarounds given in the thread you have shared, if you'd like to use the newer models available (v2020), you can build the OpenVINO™ toolkit package from source as described in their build-instruction readme available in the repository.

We are working on clear and detailed instructions on building OpenVINO™ toolkit from source for RPI/ Intel® NCS2 usage. If you decide to go this route to be able to use newer model versions, don't hesitate to share your progress and any issues you may encounter.

Best Regards,

Luis