FrozenGene commented on issue #5230: Adding support for TFLite QnnSubtract 
operator.
URL: https://github.com/apache/incubator-tvm/pull/5230#issuecomment-611330944
 
 
   > Correct. So here, TFLite converts Conv to depthwise conv, where depthwise 
conv has input_channels = groups = 1 and depth multiplier > 1
   > 
   > 
   > 
   > This PR converts this type of depthwise conv back to normal conv. Exactly 
the same way that you are suggesting. (There is not even a way to run Relay 
depthwise conv with groups = 1)
   > 
   > 
   > 
   > Let us know if that is ok.
   
   Yes. Convolution should be better I expect. However, I wish we could have 
one performance comparation result between these two ways. So that we decide 
whether we keep depthwise convolution with multiplier greater than 1 or like 
this pr converting back to normal convolution finally.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
[email protected]


With regards,
Apache Git Services

Reply via email to