[GitHub] [incubator-mxnet] yuzhms commented on issue #15570: Can MXNet support locally-connected layer?

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

yuzhms commented on issue #15570: Can MXNet support locally-connected layer?
URL: https://github.com/apache/incubator-mxnet/issues/15570#issuecomment-512632953
 
 
   It's quite simple.
   
    If we have an input with shape `(BxC_inxHxW)`，we use a convolution kernel with stride =3, and we want to output feature maps with `C_out` channels, the convolution layer will create weights with shape `(C_out, C_in, 3, 3)`. So the operation in each space position share weight.
   
   Locally connected layer operate as same as the convolution layer, except in each space position, it use different weight. So the weights now have shape of `(C_out, C_in, 3, 3, H_out, W_out)`.
   
   see[this link](https://prateekvjoshi.com/2016/04/12/understanding-locally-connected-layers-in-convolutional-neural-networks/) for more comprehensive explanation.

----------------------------------------------------------------
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