Not detecting objects when I run object_detection_demo_yolov3_async

[MBhat97]

Hello developers I am trying to do object detection using the object detection demos from build samples. 

I have downloaded the .xml and .bin files from the Attachments in the following link in :

https://software.intel.com/en-us/forums/intel-distribution-of-openvino-toolkit/topic/808504

The input is webcam, but nothing is being detected.

And also I want the input to be taken from Intel RealSense d435 camera. How do I modify the code to yield me the same?

Kindly help me with the both.

[Luis_at_Intel]

Hi Bhat, Manoj,

Thanks for reaching out. I haven't tried the link you have shared, but it looks like they were trying to use tiny-yolov2 on an older release of OpenVINO. Tiny-yolov3 should work on OpenVINO R3, there is a sample app available on the NCAPPZOO for the NCS here. I'd strongly suggest to take a look at that repo for good reference.

Couple of things you can do to have the input taken from our RealSense camera, for Python* it would look something like this:

input_stream = 0 if args.input == "cam" else args.input

    cap = cv2.VideoCapture(input_stream)
    frame = cap.read()

I believe you can change the input_stream = 0 to a different stream (like '1') in case '0' doesn't work, I tested it as is and it worked on Windows 10 and my RealSense D435. For your reference you can use the object_detection_demo_yolov3_async.py sample available under dir "C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\inference_engine\demos\python_demos\object_detection_demo_yolov3_async". Let me know if you have issues or any other questions.

Regards,

Luis

[MBhat97]

Thank you Luis_at_Intel. But I cannot find the object_detection_demo_yolov3_async.py, I am using Ubuntu 16.04, however there is directly an application file which is run when I use the command ./object_detection_demo_yolov3_async -with arguments. It would be of great help if you could attach the .py file so that I can add the code you suggested and check for inputs from Intel RealSense camera D435.

And also the code you mentioned, does it goes in def_argparser():, I am very new to computer vision, kindly help me with this.

Thanks in advance.

[Luis_at_Intel · Accepted solution]

Hi Bhat, Manoj,

I apologize! I assumed you were using Windows, the path to the sample in Linux is "/opt/intel/openvino/deployment_tools/inference_engine/demos/python_demos/object_detection_demo_yolov3_async". You can also find the sample in the Open Model Zoo repo, the Python* samples are here.

As far as the def_argparser() question goes, it depends if you expect the user to specify the argument in the command or if you want to hard code the value. For example it would be a hard code input_stream = 0, or if there is user input then you need the argument parser as previously mentioned. The sample should give you a good overview on how to go about it, let me know if you have any further questions.

Regards,

Luis

[MBhat97]

Yes thank you Luis_at_Intel I found it!!!