Forums - Feeding 4 camera inputs to C++ application for ML model

10 posts / 0 new
Last post
Feeding 4 camera inputs to C++ application for ML model
22imonreal
Join Date: 10 Feb 21
Posts: 80
Posted: Thu, 2021-07-01 02:54

I have seen different options, but what is the best way to get 4 camera feeds into a C++ application for postprocessing and then feeding them to a ML model in the RB5?

And would you be able to provide examples to illustrate your suggestions?

Thanks in advance.

  • Up0
  • Down0
kevin.dai
Join Date: 21 Oct 20
Posts: 137
Posted: Wed, 2021-07-07 14:13

Hi 22imonreal,

Let me check with SNPE application engineers and then give your suggestion 

Thanks

Kevin

 

  • Up0
  • Down0
kevin.dai
Join Date: 21 Oct 20
Posts: 137
Posted: Thu, 2021-07-08 11:20
You can do in these steps:
 
1. Set all 4 cameras resolution ratio
2. Store the 4 cameras output tensor to one USERBUFFER tensor with 4 batch size.
3. Execute the network with 4 batch sizes and get the result from the output USERBUFFER
Notice that USERBUFFER is recommanded as it has a better performance.
  • Up0
  • Down0
22imonreal
Join Date: 10 Feb 21
Posts: 80
Posted: Tue, 2021-07-13 01:47

Hello, and thank you for your answer.

Would you mind provinding an example of how I would go about doing each of the steps you mentioned above?

Example commands, or example projects or scripts, any hints would be very much appreciated.

Thank you in advance.

Kind regards.

  • Up0
  • Down0
kevin.dai
Join Date: 21 Oct 20
Posts: 137
Posted: Wed, 2021-07-14 15:12

Hi 22imonreal

There are not exact sample code for 4 cameras stream input for SNPE network application in SNPE SDK.

The  steps are suggestion for you to develop your application. 

 

Thanks

Kevin

  • Up0
  • Down0
22imonreal
Join Date: 10 Feb 21
Posts: 80
Posted: Thu, 2021-08-05 08:54

Hello,

Would you be able to provide some code showing how store the 4 cameras output tensor to one USERBUFFER tensor with 4 batch size?

Thanks

  • Up0
  • Down0
mitarath
Join Date: 21 Jun 21
Posts: 12
Posted: Wed, 2021-08-18 16:14

Hi Csutomer,
we do not have smaple code for above use case and RB5 does not support opening 4 cameras at the same time.

Regards,
Mita

  • Up0
  • Down0
22imonreal
Join Date: 10 Feb 21
Posts: 80
Posted: Thu, 2021-08-19 03:25

Okay, but how would I store the 4 (or however many the RB5 can open simultaneously) cameras output tensor to one USERBUFFER tensor with 4 batch size?

  • Up0
  • Down0
22imonreal
Join Date: 10 Feb 21
Posts: 80
Posted: Mon, 2021-08-23 07:35

PLEASE, DON'T IGNORE MY QUESTIONS (SEE ABOVE)

  • Up0
  • Down0
mita.rathod
Join Date: 17 Sep 18
Posts: 59
Posted: Tue, 2021-09-07 14:17

hello customer,

This parallel method is available on our device. There is an API in snpe tools for testing.If you downloaded the snpe source code, you can find this API in this directory /doc/html/tools.html snpe-throughput-net-run.

  • Up0
  • Down0
or Register

Opinions expressed in the content posted here are the personal opinions of the original authors, and do not necessarily reflect those of Qualcomm Incorporated or its subsidiaries (“Qualcomm”). The content is provided for informational purposes only and is not meant to be an endorsement or representation by Qualcomm or any other party. This site may also provide links or references to non-Qualcomm sites and resources. Qualcomm makes no representations, warranties, or other commitments whatsoever about any non-Qualcomm sites or third-party resources that may be referenced, accessible from, or linked to this site.