Loading detection models and reusing (based on multi_camera_multi_person demo)

[Zvi]

I would like to load a person detector once, use it on one clip and then reuse the model on another clip (based on the MultiCameraTracker demo). 

When I try to do this I get an error :

ie_api.pyx in openvino.inference_engine.ie_api.ExecutableNetwork.start_async()

ValueError: Incorrect request_id specified!

I created a python notebook that recreates the problem. Attached.

Thanks

[Adli]

Hi Zvi,

Thank you for reaching out to us. I assume that the error is triggered possibly due to request_id is holding information from the previous clip. Try resetting the demo before attempting to execute it on a new clip.

Regards, 

Adli

[Zvi]

Thanks, but this is exactly my question.

object_detector = multi_camera_multi_target_tracking.Detector(ie, detector,config['obj_det']['trg_classes'],t_detector,detector_device, cpu_extension,1)

 If this is the object detector - this is where the model is loaded. I want to be do this once. How do I reset this object for a different clip (run)?

[Adli]

Hi Zvi,

If possible, could you share your program (model, file) here for replicating purposes?

Regards,

Adli

[Zvi]

Hi,

It's attached to the original question (Python notebook to reproduce problem) - but attaching again. 

Thanks

[Adli]

Hi Zvi,

We are currently investigating this issue and will get back to you soon.

Regards,

Adli

[Adli]

Hi Zvi,

Thank you for your patience. The 'Multi-Camera Multi-Target Tracking' sample is not designed for reusing the same object_detector variable with different videos.

In order to make your Jupyter Notebook sample code work, you could just reinitialize the 'object_detector' before calling out the 'my_run' function for the second time. So the last section of the code would look like this:

capture = multi_camera_multi_target_tracking.MulticamCapture([clip_name])
object_detector = multi_camera_multi_target_tracking.Detector(ie, detector,config['obj_det']['trg_classes'],t_detector,detector_device, cpu_extension,1)
my_run(params, config, capture, object_detector, object_recognizer)

Regards,

Adli

[Zvi]

Hi,

Thanks you for your answer.

Could you explain the design issues about the problem of reusing the same object (that is missing in the sample)? Some documentation? What would I need to change? A snippet?  

Is this the same for any model that uses IECore? 

I know I could reload the model again and rerun but that is exactly what I wish to prevent. 

[Adli]

Hi Zvi,

We are currently investigating this issue and will get back to you soon.

Regards,

Adli

[Adli]

Hi Zvi,

I will close this thread since the issue was solved based on the following Github thread: https://github.com/openvinotoolkit/openvino/issues/3039

Regards,

Adli

[Adli]

Hi Zvi,

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

Regards,

Adli