Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Mirko
Novice
520 Views

C++ Object detection demo yolov3 async Crashes

Hi all,

I already ran the human_pose_estimation_demo and integrated it as part of my project.

I would like to do the same with the object detection trained by yolov3.

I followed the link below:

https://docs.openvinotoolkit.org/latest/omz_demos_object_detection_demo_yolov3_async_README.html

In  https://download.01.org/opencv/ link I didn't find any pretrained models for yolov3 (there are modules only for yolov2) .

I followed exactly the instruction at  the link https://docs.openvinotoolkit.org/latest/openvino_docs_MO_DG_prepare_model_convert_model_tf_specific_... . I haven't got any errors, but, when I use the  .xml and .bin files created, the demo crashes openVino1.jpg

It seems to be a problem about the accuracy of the IR model generated. As if the ngraph::Function's describing the model were malformed or cnnNetwork object wasn't able to get the ngraph::Functions from the model.

I had this issue some months ago using the 2020.3 version and I have hoped that switching to 2020.4 version and restarting from scratch could have corrected any possible mistake I did or any bugs. Every step have succeded but the demo is still not running.

I have also tried to deal with the json file and some parameters in the IR model creation steps , but nothing has changed.

Has anyone a IR model working with this demo? 

Has anyone any suggestion to solve this problem?

Thank you 

Mirko

 

0 Kudos
12 Replies
Mirko
Novice
518 Views

Sorry I didn't note the Visual was set in Italian language.

The error showed is an Unhandled Exception: Access Violation

Mirko

Iffa_Intel
Moderator
495 Views

Greetings,


If you receive an Unhandled Exception Access Violation error when you run an application on your Windows 10/8/7 computer, it probably means that some part of the program code tried to access a protected memory address and was denied access.


Probably you can try to run the Visual Studio as admin.


Sincerely,

Iffa


Mirko
Novice
492 Views

Hi Iffa,

I didn't mentioned it, but I ran Visual as administrator. I always do that.

Any other suggestion?

Thank you

Mirko

Iffa_Intel
Moderator
480 Views

Sometimes the location such as in windows "Program Files x86" has problems when it's called.

Whenever including this kind of path ensure to add "" with it.

for instance: python mo_ty.py --input_model C:/"Program Files(x86)"/IntelSWTools/openvino_2020.4 ...


Sincerely,

Iffa


Mirko
Novice
472 Views

Hello Iffa,

sorry but I think it isn't a file access violation. I think the file is opened and the network is loaded.

As you can see after auto cnnNetwork = ie.ReadNetwork(FLAGS_m); some parameters are correctly set in cnnNetowork and in some line code them are retrived . (input and output info)

OV2.jpgOV1.jpg

So I think we can say the loading network step succeed.

It seems the netwrok is not consistent in all its parameter values. The extraction of the yolo regions fails.

The xml e bin files are in a folder in C:/Users/... (as you can see below) and I also tried to run the example from  the Command Prompt (opened as Administrator).  I have added some logs

OV4.jpg

and them show that the code stops working at the same point:

OV3.jpg

 

I really need a way forward to overcome this issue.

Best regards

Mirko

Iffa_Intel
Moderator
460 Views

Greetings,

 

To ensure that you are using the correct steps and the correct model, you can refer to this video as this will give you clearer idea: https://www.youtube.com/watch?v=FaqVhvJ6-Uc

 

Steps for yolov2 and yolov3 is quite different. I had done a few experiment myself with yolov3 with Openvino 2020.4 and I never get this kind of error.

 

Try to get the same samples as per video and check whether you could run it.

If you didn't want to use the same as video(Anaconda) you can use the same model as he's using and manually run the command as per in official documentation: https://docs.openvinotoolkit.org/latest/openvino_docs_MO_DG_prepare_model_convert_model_tf_specific_...

 

Sincerely,

Iffa

 

 

Iffa_Intel
Moderator
444 Views

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


Mirko
Novice
435 Views

Sorry Iffa, 

but i haven't had time to try your hint until yesterday. THe solution doesn't work. I followed the video using Jupyter, but the script execute the instruction i have already followed...and the secondo link you shared is the same I followed the first time as wrote in my original post. So I think the issue is not closed.

How can I go on with this question? Do I have to open a new ticket?

Best regards

Mirko

BSung8
Novice
411 Views

Try Release mode.

Mirko
Novice
393 Views

Than you BSung8,

actually in release it works. Do you have any hint about the reason?

Are you working on the object detection with weights imported from yolov3 weights?

Mirko

 
 
BSung8
Novice
363 Views

I think that's /debug/ngraph.dll bug. 

And I duplicate this bug in object_detection_demo_ssd_async/object_detection_demo_yolov3_async intel demo code.

 

Mirko
Novice
351 Views

Hi,

thank you for your response. 

What do you mean with "I duplicate this bug in object_detection_demo_ssd_async/object_detection_demo_yolov3_async intel demo code." ?

BR

Mirko

Reply