[GitHub] anirudhacharya commented on issue #11281: LRN operator doesn't comply with ONNX specification

[GitBox <gi...@apache.org>]

anirudhacharya commented on issue #11281: LRN operator doesn't comply with ONNX specification
URL: https://github.com/apache/incubator-mxnet/issues/11281#issuecomment-397474431
 
 
   @matteosal LRN indeed does accept only odd values for the window size, the LRN operator introduced in [this paper](https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf) also prescribes the same. If the window size is even, then the local region over which normalization is performed will not be symmetric over the given element and will lead to erroneous results. 
   
   If you look at the ONNX's test and example code, only odd number window sizes are being tested for - https://github.com/onnx/onnx/blob/master/docs/Operators.md#examples-31 It is a bug on the ONNX specification side that even valued window size is not being raised as an error. I will take it up with the ONNX community, and fix it.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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