[OpenVINO][HDDLPlugin][Windows 10] running multiple models failed when on VPU alone

[Susanto__Jeffry]

Hi, I'm currently working with a Movidius™ Myriad™ X on Windows 10 (device is VPC-3350s). I followed the installation guide here to install the latest OpenVINO 2019 R3.

However, when I run demo_security_barrier_camera.bat -d HDDL , some warnings suddenly appeared:

[16:47:57.8235][4656]W[ConfigParser.cpp:269] Warning: Cannot find key, path=scheduler_config.max_graph_per_device subclass=0, use default value: 1.
[16:47:57.8247][4656]W[ConfigParser.cpp:292] Warning: Cannot find key, path=scheduler_config.use_sgad_by_default subclass=0, use default value: false.

and also after closing the picture which is stuck at 0 fps,

 0.0FPS for (2 / 1) frames

Detection InferRequests usage: 0.0%

This problem occurs if the HDDL plugin is running multiple models (same thing happened with python face recognition demo running on HDDL). This problem doesn't occur when I ran classification_sample_async.exe on HDDL.

With the GPU/CPU everything goes well, however I am hoping to be able to utilize the VPU.

I am wondering if anybody have encountered this problem and how to solve it. (I have followed the steps in the installation guide, including converting to FP16, installing the .inf, etc, but this still happened)

Thank you very much.

Jeffry

[Morabito__Roberto]

Hi,

I'm facing the same problem. Basically, with any of the demo I'm able to run multiple models on the VPU.

Did you figure out how to solve the issue?

The interesting thing is that if I run simultaneously two different demo and allocate one model each to the VPU or if, by modifying the code, I run the same model twice or more on the same VPU, everything goes smooth. 

Mysterious behaviour. 

Roberto