Forums - For the same input data and the same dlc file, we get different results on different devices.

2 posts / 0 new
or Register
Last post
For the same input data and the same dlc file, we get different results on different devices.
songmf
Join Date: 22 Dec 21
Posts: 1
Posted: Wed, 2021-12-29 09:43
Top

I tried to run the inference of our network with snpe-net-run on different devices, on CPU mode.

I tried four devices. The first one is PC, the second one is snapdragon 636, the third one is snapdragon 865, the fourth one is snapdragon 888. I have the same result on PC,  snapdragon 636 and snapdragon 888. But the result on snapdragon 865 is very different. And I also check the output of every layer by adding --debug. I found the result of the first layer is not right. 

Our first layer is as following:

self.first_conv = nn.Sequential(

            nn.Conv2d(3, 24, 3, 2, 1, bias=False),

            nn.BatchNorm2d(24),
            nn.ReLU(inplace=True),
        )
There are totally 98304 values in the output tensor, and the first 88512 values are the same accross all devices. And the remaining values on snapdragon 865 is different from the results on other devices.
 
Can anyone tell me how to solve the problem?  Best regards. Since our application should be deployed on snapdragon 865, we should get correct results on this device.
 
  • Up0
  • Down0
Re: For the same input data and the same dlc file, we get...
weihuan
Join Date: 12 Apr 20
Posts: 270
Posted: Thu, 2021-12-30 17:48
Top

Dear customer,

What's target you implmented for the platform you mentioned? It works well if model accuracy exists little difference if run on different hardware. The X86 , ARM, DSP consisits of different hareware architecture. 

BR.

Wei

  • 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.