Forums - Same model structure, different input shape, one can get correct result the other cannot!

2 posts / 0 new
or Register
Last post
Same model structure, different input shape, one can get correct result the other cannot!
383736929
Join Date: 17 Oct 18
Posts: 2
Posted: Mon, 2019-05-13 17:52
Top

Hi all,

I have two models, both of which network structure are the same only except the input tensor shape, one is {1,512,512,6} and the other is {1,512,512,9}. The model of input tensor shape {1,512,512,6} can get the correct result running on android. However the other model, {1,512,512,9}, 

cannot get the correct result. These two models are exactly the same, which only difference is the input shape. So I am very confused why this happen. 

 -We tested the various SNPE version, 1.23.1, 1.24 and 1.25,1 . These SNPE all had the same problem that the model result of input {1,512,512,6} shape is correct, but the other model reuslt was not correct when we increased the 4th-d from 6 to 9.

-The both models coverted from the tensorflow pb model, and their pb models both can get the correct result.

-There also weren't any error or warning, when we converted the two pb models to DLC models.

Are there some bugs in the SNPE lib on android?

Can anyone help me?

 

  • Up0
  • Down0
Re: Same model structure, different input shape, one can get...
Vihari
Join Date: 20 May 19
Posts: 1
Posted: Mon, 2019-05-20 06:58
Top

Hi, 

May I know which model you are trying? So that, if possible I can try once and come back to you on the same.. It would be helpful to analyse in a better way.

Thank you

  • 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.