You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@systemml.apache.org by "Mike Dusenberry (JIRA)" <ji...@apache.org> on 2017/06/28 20:00:01 UTC
[jira] [Updated] (SYSTEMML-1516) Improve output size calculation in
conv2d & max_pool2d
[ https://issues.apache.org/jira/browse/SYSTEMML-1516?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Mike Dusenberry updated SYSTEMML-1516:
--------------------------------------
Issue Type: Improvement (was: Sub-task)
Parent: (was: SYSTEMML-618)
> Improve output size calculation in conv2d & max_pool2d
> ------------------------------------------------------
>
> Key: SYSTEMML-1516
> URL: https://issues.apache.org/jira/browse/SYSTEMML-1516
> Project: SystemML
> Issue Type: Improvement
> Reporter: Mike Dusenberry
> Assignee: Mike Dusenberry
> Priority: Minor
> Labels: SYSTEMML-618
> Fix For: SystemML 0.14
>
>
> The calculation of the output sizes (`Hout`, `Wout`) for conv2d and
> max_pool2d layers rely on a flooring operation to only allow for valid
> operations. Currently, we simply rely on the `as.integer` call to
> implicitly perform this flooring operation during truncation of the double
> to an integer. While this gives the correct results, it is less
> readable than it should be, and relies on truncation behavior than explicit mathematical behavior. We should simply add an explicit `floor`
> operation to each calculation.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)