Issues while loading model detector files [VS2017, Windows]

[galligio]

Hello everyone,

I am trying to build an efficient face detection system for my real-time Visual C++ application which uses an Intel Realsense camera as input stream. For this purpose, I've tried to refer to these two samples: this one from the librealsense repository and this one from the OpenViNO toolkit in order to be able to both detect the faces and to estimate the head pose.

Here are some information about my developing environment:

- Windows 10

- Intel Realsense D435

- OpenViNO toolkit 2019.R3 (provided by the Intel Realsense SDK 2.0) with OpenCV 4.1.2 (Note that I've also tried other OpenViNO versions but I couldn't solve the issue...)

After making some small code changes to these files, in order to match some functions, I've managed to successfully build my VS solution and run it in Debug mode. However, when trying to run it in Release mode, the program crashes at the following line:

 faceDetector.load_into(engine, device_name);

In particular, it crashes when trying to load the network with LoadNetwork (last line of the load_into function from the base-detection.cpp file).

Am I missing something in the Release configuration? The

PATH=<INSTALL_DIR>\deployment_tools\inference_engine\bin\intel64\Release;

PATH variable in the Environment field is already set.

Other attempts I've made and related issues:

• I've tried to build OpenViNO 2020.1 from the open source repo and to include it in my project (in order to run it correctly, I had to comment out some small code parts from the win_shared_object_loader.h file, see the issue below in the Realsense community for more details). When trying to load the .xml model, the app crashes and I get the following log:

[INFO] Loading object detection model to the CPU device
[INFO] Loading object detection model from: models/face-detection-retail-0004/face-detection-retail-0004.xml
[ERROR] Failed to load face detector model files:
Cannot create ShapeOf layer fc7_mbox_priorbox/0_port id:192

• With OpenViNO 2021.1 built from source code, these are the errors I get:

[INFO] Loading object detection model to the CPU device
[INFO] Loading object detection model from: models/face-detection-retail-0004/face-detection-retail-0004.xml
[ERROR] Failed to load face detector model files:
invalid stoi argument
[INFO] Loading Head Pose model to the CPU device
[INFO] Loading network files for Head Pose Estimation network
[ERROR] Failed to load head pose estimator model files:
Cannot create Gather layer angle_p_fc/weights_transpose/Shape/Gather id:138 from unsupported opset: opset7
C:\j\workspace\private-ci\ie\build-windows-vs2019@2\b\repos\openvino\inference-engine\src\readers\ir_reader\ie_ir_parser.cpp:594

I've posted this issue in Support Realsense community and I got redirected here. If anyone knows any possible solution to this, I would very much appreciate your help.

Thank you

[IntelSupport · Accepted solution]

Hi galligio,

Thanks for reaching out. For the crashed when load network issues, do you install OpenVINO from source or from the OpenVINO installation package? Make sure in your <INSTALL_DIR>\deployment_tools\inference_engine\lib\intel64\Release directory, you have these 3 file: inference_engine.lib, inference_engine_c_api.lib, inference_engine_transformations.lib

Furthermore, the "Cannot create ShapeOf layer fc7_mbox_priorbox/0_port id:192" error is due to the unsupported network. For your information IE Plugin and IE Network have been deprecated. I would recommend using the IE Core. Therefore, It is recommended to use the latest version of the OpenVINO toolkit which is the OpenVINO 2021.4. The latest version is more stable and upgraded.

From your opset error, it looks like you use the old version of operators and it is unsupported for 2020.1 OpenVINO. Can you share your workaround and command for this? Are you running any OpenVINO samples/demos? Are you using the model from 2020.1 or from the previous version?

Regards,

Aznie

[galligio]

Hi Aznie,

thank you very much for the response.

• For the crashed when load_network issues, i can confirm that the files you mentioned are present

• About the suggestion to use OpenViNO 2021.4, does it not require Visual Studio 2019? Unfortunately I need to use Visual Studio 2017 for other reasons. Can I use OpenViNO 2021.4 anyway? Or should I use version 2021.1/2 instead?

• I am now running demos/samples successfully

Regards,

gg

[IntelSupport]

Hi galligio,

OpenVINO 2021.4 is compatible with Visual Studio 2019 only. The support for Visual Studio 2017 with the OpenVINO toolkit is only valid until OpenVINO 2021.2 version.

Regards,

Aznie

[galligio]

Hi Aznie,

I have now managed to run my application with OpenViNO 2021.1 toolkit, installed from the installation package, in both Debug and Release modes. I was indeed loading a previous version of the models, which caused the opset error.

Thank you very much for all the help

Best regards,

gg

[IntelSupport]

Hi galligio,

I am glad to hear that you are able to run your application now. This thread will no longer be monitored since this issue has been resolved. If you need any additional information from Intel, please submit a new question. 

Regards,

Aznie