Loading multiple model in a single NCS2

[hmc_roboticslab]

Hello,

sorry for consecutive question in a row.

We also are trying to load multiple light-weight networks on a single NCS2.

We use a following function of IECore instance to search available NCS2 devices.

self.ie.get_metric("MYRIAD", "AVAILABLE_DEVICES")

which returns 2 devices ['1.11.1', '1.11.2']

However, after loading a model on one of two devices (e.g., '1.11.1') in a returned list using "self.ie.load_network()",

the device disappears from available device list in next call (e.g., ['1.11.2']).

So, we failed to load multiple models in same device using this method.

Here are my questions,

1. Is there any recommended way to load multiple models in a same NCS2 device?

2. Could openvino.runtime (openvino API 2.0) be better solution compared to inference_engine API for our purpose?

Thanks in advance,

[Iffa_Intel · Accepted solution]

Greetings,

The OpenVINO Open Model Zoo Demo does have a few samples that use multiple networks/models in a single inference.

One of them is this Action Recognition Python Demo. You may refer to this demo to see how they are handled.

The latest API 2.0 might be better if it's convenient for you to upgrade your current version (you don't have any specific dependencies on the current version you are using).

Sincerely,

Iffa

[hmc_roboticslab]

Sorry of late response.

We successfully adopted multi-model in a single NCS2 referring linked example code.

The biggest mistake was that we created Core instance (IECore in openvino 1.0) for each algorithm.

Core instance should be created once and shared to load different models on a same device.

Thanks for solution!

[Iffa_Intel]

Greetings,

Intel will no longer monitor this thread since we have provided a solution. If you need any additional information from Intel, please submit a new question. 

Sincerely,

Iffa