Laurawly commented on pull request #6108:
URL: https://github.com/apache/incubator-tvm/pull/6108#issuecomment-674562133


   > @trevor-m @yongwww @Laurawly
   > For the `get_valid_counts` part, I misunderstand it because I didn't 
understand `argsort` correctly and I have recovered to the original version, 
which is much faster.
   > 
   > For the `rearrange_indices_out` part, which is necessary because the 
result of `nms` is used by a `strided_slice` in `def _nms` in tensorflow 
frontent, I agree that the current way may regress the performance, but since 
we need to do data arrangement in this function, I can hardly figure out a 
better way to implement it.
   
   Could you show some benchmark numbers regarding the changes? @yongwww could 
have a better comment on the tensorflow related changes. Also it seems that 
there's illegal memory access error based on CI. 


----------------------------------------------------------------
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]


Reply via email to