Up AI Core doesn't work well with OpenVINO Toolkit

[Lu__Gavin]

Hi there,

I am trying to make my Up AI Core work with OpenVINO Toolkit, but met with some issues.

This is a small box with Atom x5-E3940, 1.6GHz (4 Cores) & 12 Execution Units and 8GB RAM, and I installed Ubuntu 16.04 there.

I got Up AI Core hardware detected at boot time with dmesg as below:

I followed closely with guidance at https://software.intel.com/en-us/articles/OpenVINO-Install-Linux to enable user access to the mPCIe card before reboot the box.

Then I executed ./demo_security_barrier_camera.sh -d MYRIAD in folder opt/intel/computer_vision_sdk/deployment_tools/demo/. To be noticed that I executed same cmd with parameters -d GPU/CPU and got the result well already before I tried MYRIAD.

And I got error output like this:

I fixed it after I modified file demo_security_barrier_camera.sh in folder /opt/intel/computer_vision_sdk_fpga_2018.3.343/deployment_tools/demo.

The last few lines are to call functions built to run inference in folder FP32. Just change these positions from FP32 to FP16, which were also built out earlier in model optimization. Then that is done.

But is that expected, or did anymore meet this issue before? I assumed that Up AI Core shares the same IP core with Myriad Neural stick but just difference on physical interfaces (USB v.s. mPCIe).

[Lu__Gavin]

Please refer to https://forum.up-community.org/discussion/3307/up-ai-core-doesnt-work-well-with-openvino-toolkit as there is some issue on error message format here.

[Monique_J_Intel]

Hi Gavin,

Issue here is the demo script has the paths of the vehicle license plate barrier detection, vehicle attributes recognition, and license plate recognition models are hardcoded to use FP32 versions of the .xml which is only suitable to run on CPU as MYRIAD and GPU require FP16. This bug has been fixed in the upcoming release R4 of OpenVINO.

Kind Regards,

Monique Jones

[Lu__Gavin]

Glad to know that!