[jira] [Commented] (SINGA-51) Use Caffe's im2col and pooling implementation to accelerate conv and pooling operations

["ASF subversion and git services (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SINGA-51?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14743302#comment-14743302 ] 

ASF subversion and git services commented on SINGA-51:
------------------------------------------------------

Commit e7691422c1c09b3ebea5ff399be1724ec72a5c9d in incubator-singa's branch refs/heads/master from wang sheng
[ https://git-wip-us.apache.org/repos/asf?p=incubator-singa.git;h=e769142 ]

SINGA-51 Improve the convolution and pooling operations

merge with pull request #64


> Use Caffe's im2col and pooling implementation to accelerate conv and pooling operations
> ---------------------------------------------------------------------------------------
>
>                 Key: SINGA-51
>                 URL: https://issues.apache.org/jira/browse/SINGA-51
>             Project: Singa
>          Issue Type: Improvement
>            Reporter: wangwei
>
> Currently SINGA uses [Mshadow|https://github.com/dmlc/mshadow/tree/master/doc]'s unpatch2col and pool expression in the implementation of convolution and pooling layer. After comparing them with Caffe's implementation using im2col and pooling implementation, we found the later implementation is faster. Hence we can add them into SINGA as two new layers, CaffeConvLayer and CaffePoolingLayer.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)