tvm-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From GitBox <...@apache.org>
Subject [GitHub] [incubator-tvm] FrozenGene edited a comment on issue #5230: Adding support for TFLite QnnSubtract operator.
Date Thu, 09 Apr 2020 03:40:31 GMT
FrozenGene edited a comment on issue #5230: Adding support for TFLite QnnSubtract operator.
URL: https://github.com/apache/incubator-tvm/pull/5230#issuecomment-611310639
 
 
   > Thanks for the reply. Can you please share what two things we are comparing? I am
still confused. Please let me know example of input shape, depth multiplier, groups etc.
   
   I leverage this issue https://github.com/tensorflow/tensorflow/issues/23563 we talked about.
For example we feed the input (1, 224, 224, 1)´╝îin our tensorflow we have convolution to
handle it. i.e. `input -> convolution`. However, the tflite will change the convolution
into depthwise convolution with multiplier greater than 1, the input channel will be 1.  Like
this picture of tensorflow issue:
   ![image](https://user-images.githubusercontent.com/7287321/78855497-a5766f80-7a56-11ea-92d3-96590ffb4fae.png)
   I suspect if we don't change to depthwise convolution with multiplier greater than 1 and
keep the original convolution, like this:
   ![image](https://user-images.githubusercontent.com/7287321/78855555-c8088880-7a56-11ea-8fe7-c4bb0183a998.png)
   We could get better performance.
   

----------------------------------------------------------------
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:
users@infra.apache.org


With regards,
Apache Git Services

Mime
View raw message