systemml-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Prithviraj Sen (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SYSTEMML-1483) Add Deconvolution layer in nn library and Caffe2DML
Date Wed, 03 May 2017 23:22:04 GMT

    [ https://issues.apache.org/jira/browse/SYSTEMML-1483?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15995894#comment-15995894
] 

Prithviraj Sen commented on SYSTEMML-1483:
------------------------------------------

I tried running conv2d_backward_data with a 3X3 filter and 2X2 delta map (no padding, stride=1).
It returned a 3X3 output. Running a valid convolution on a 3X3 input with a 3X3 filter does
not return a 2X2 output. That's why I couldn't figure out whether the behavior of conv2d_backward_data
was correct, in this case.

Running a valid convolution with a 3X3 filter on a 4X4 input does return a 2X2 output. This
is why, I was expecting a 4X4 result in the previous case.

> Add Deconvolution layer in nn library and Caffe2DML
> ---------------------------------------------------
>
>                 Key: SYSTEMML-1483
>                 URL: https://issues.apache.org/jira/browse/SYSTEMML-1483
>             Project: SystemML
>          Issue Type: Sub-task
>            Reporter: Niketan Pansare
>
> http://caffe.berkeleyvision.org/tutorial/layers/deconvolution.html
> [~mwdusenb@us.ibm.com] [~prithvi_r_s] 



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Mime
View raw message