Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.
6415 Discussions

multistick plugin doesn't working on face_recognition_demo.py

glennford49
Novice
884 Views

Good day to all,
I have successfully run face_recognition_demo.py on my notebook using ncs2, I have achieved 11FPS on a single stick and decided to double its FPS to achieve real time detection.However using 2 NCS2,
with "MULTI:MYRIAD.3.2.3-ma2480,MYRIAD.3.2.1-ma2480" as my default device doesn't do any significant increase of my FPS.
I have only added a single line on face_recognition_demo.py with
multMyriad ="MULTI:MYRIAD.3.2.3-ma2480,MYRIAD.3.2.4-ma2480"
and changing all the default "CPU" to multMyriad
Did I missed something in here?
O.S = ubuntu 20.14
openvino env = openvino 2021.2.185
demo = face_recognition_demo.py

0 Kudos
6 Replies
Peh_Intel
Moderator
863 Views

Hi Glenn,

Thanks for reaching out to us.

To increase FPS using multiple NCS2 devices, I suggest you create multiple inference request for each ExecutableNetwork. These requests can be processed by the device with a level of parallelization that best works with the target devices. For Intel® NCS 2 devices, four inference requests for each Executable Network are recommended for optimum performance.

These information are available in the following article (in the 'Multiple NCS Devices' section):

https://software.intel.com/content/www/us/en/develop/articles/transitioning-from-intel-movidius-neural-compute-sdk-to-openvino-toolkit.html

 

I share with you the following link, which has an example implementation:

https://github.com/yas-sim/openvino-multi-ncs2-throughput-mode

 

Regards,

Yu Chern

 

 

0 Kudos
glennford49
Novice
849 Views

Thanks for your response.
https://github.com/yas-sim/openvino-multi-ncs2-throughput-mode
I have tested the link above,running the multi-ncs.py clearly double the FPS,
However when I modify it to use the cv2.VideoCapture().
Loading the model person-detection-retail-0013.
execnet.requests[reqId].async_infer(inputs={inblob:frame}) # Asynchronous inference
returns None
and
execnet.infer(inputs={inblob:frame}) # Synchronous inference
return an array
My only choice for the script to work is using the Synchronous inference stated above.
Single stick yields 6 FPS , TWO sticks yields the same 6 FPS.
With this I cant measure the fps of my two ncs2 sticks.

0 Kudos
Peh_Intel
Moderator
809 Views

Hi Glenn,

Greetings to you. I apologize for the late response. Thanks for your time trying the demo shared before.

Performance of MULTI device depends on the application actually. For synchronous mode inferencing, MULTI doesn’t improve performance (since 2 devices don't execute the same request).

For asynchronous mode inferencing, MULTI improves performance (when 2 devices execute 2 different requests) and it should double the FPS (in case of 2 NCS2 sticks).


Also, for your information, MULTI plugin seems to not be supported for the Interactive Face Recognition Demo, as can be seen from the README file.


Regards,

Yu Chern


0 Kudos
Peh_Intel
Moderator
781 Views

Hi Glenn,


Thank you for your question. If you need any additional information from Intel, please submit a new question as this thread is no longer being monitored.

 

 

Regards,

Yu Chern


0 Kudos
RobertCrook
Beginner
771 Views

Try increase FPS using multiple NCS2 devices.

0 Kudos
RobertCrook
Beginner
745 Views
0 Kudos
Reply