Forums - Does SNPE supports neural network with non-image data

4 posts / 0 new
Last post
Does SNPE supports neural network with non-image data
wmq387
Join Date: 7 Aug 17
Posts: 1
Posted: Tue, 2017-08-08 09:07

As in the reference guide, the input is defined as image. Is SNPE applicable to train a CNN with non-image data that is structured and localized? Thank you.

  • Up0
  • Down0
moljaca moderator
Join Date: 25 Jul 17
Location: San Diego
Posts: 40
Posted: Tue, 2017-08-08 16:41

Hi,

Thank you for interest in Snapdragon NPE. 

The reference guide is using image classification use case just as an example, because image processing is one of dominant use cases for CNNs. SNPE is inference (i.e. forward propagate) engine only and is agnostic to type of input data. As long as a CNN model can be converted into DLC, such CNN can be loaded and excuted in SNPE by feeding input data in the format CNN was designed for and trained with. Preparing input data is at user application level.

Hope this helps.

Best regards.

Thanks

  • Up0
  • Down0
22imonreal
Join Date: 10 Feb 21
Posts: 80
Posted: Wed, 2021-05-12 07:14

Hello,

What file formats does the snpe-dlc-quantize --input_list accept as inputs?

The example shows: https://developer.qualcomm.com/docs/snpe/tools.html#tools_snpe-net-run_i...

Input_1:=Placeholder_1/real_input_inputs_1/0-0#e6fb51.rawtensor

 

I want to input a (7,1) vector with values in the range {0,1}. How would I write that input in a suitable file format to be fed into the --input_list arguments?
 

Thank you in advance.

  • Up0
  • Down0
22imonreal
Join Date: 10 Feb 21
Posts: 80
Posted: Wed, 2021-05-26 01:45

Hello,

I was able to feed a 1D vector input in a txt file. However, I now need to feed in a 6 channel image (a concatenation of 2 RGB images).

Since raw images are a 3 channel GBR image and txt files can only represent 2D vectors when written with np.savetxt, how would I represent this 6 channel image in a txt file to be fed to the model with snpe-net-run?

Thanks,

  • 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.