Overview

There is a tutorial on Python API code in OpenVINO using Raspberry Pi and Neural Compute Stick on the here site, but if you want to do it in a PC-only environment, I need to make some changes and additions, so I will write a note of it. Image source: Deep learning inference learned from scratch with OpenVINO ™

The following three items are required

-Change target device -Changed FP16 to FP32 -Add library load

Change target device

Just change from MYRAID, which means Neural Compute Stick, to CPU. This is easy.

#Change before
plugin = IEPlugin(device="MYRIAD")

#After change
plugin = IEPlugin(device="CPU")

Change FP16 to FP32

Change the trained model (xml and bin files) from FP16 to FP32 For example, face-detection-retail-0004 for OpenVINO 2019_R3.1 can be downloaded from FP32 on the following site. Intel Open Source Technology Center

By the way, it may work with FP16 as it is, but FP32 is recommended, so it is better to change it. Image source: Intel OpenVINO ™ TOOLKIT official website

Add library load

Depending on the trained model you use, you may get the following errors (such as face recognition):

Traceback (most recent call last):
  File "XXXXXXXX.py", line XXX, in <module>
    exec_net = plugin.load(network=net)
  File "ie_api.pyx", line 547, in openvino.inference_engine.ie_api.IEPlugin.load
  File "ie_api.pyx", line 557, in openvino.inference_engine.ie_api.IEPlugin.load
RuntimeError: Unsupported primitive of type: PriorBoxClustered name: fc7_mbox_priorbox

As a response, it is necessary to add CPU Extensions library reading. The path and file name differ depending on the OS and CPU, so please try various things.

Example of Windows 10 (Intel Core processor)

#Add under target device specification
plugin.add_cpu_extension("C:/Program Files (x86)/IntelSWTools/openvino/deployment_tools/inference_engine/bin/intel64/Release/cpu_extension_avx2.dll")

If the above does not work, search for cpu_extension.dll and specify the full path.

Linux (Intel Atom Processor) Example

#Add under target device specification
plugin.add_cpu_extension('/opt/intel/openvino/deployment_tools/inference_engine/lib/intel64/libcpu_extension_sse4.so')

For Intel Core processors, try replacing it with libcpu_extension_avx2.so.

Mac example

#Add under target device specification
plugin.add_cpu_extension('/opt/intel/openvino/deployment_tools/inference_engine/lib/intel64/libcpu_extension.dylib')

With this, if you have a PC, you can make deep learning inference for free! see you.

OpenVINO using Inference Engine Python API in PC environment