You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@singa.apache.org by "ASF subversion and git services (JIRA)" <ji...@apache.org> on 2016/10/21 15:33:58 UTC
[jira] [Commented] (SINGA-267) Add spatial mode in batch
normalization layer
[ https://issues.apache.org/jira/browse/SINGA-267?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15595439#comment-15595439 ]
ASF subversion and git services commented on SINGA-267:
-------------------------------------------------------
Commit 61faa840e155e4259920141c3c909483fa1c14f2 in incubator-singa's branch refs/heads/master from WANG Ji
[ https://git-wip-us.apache.org/repos/asf?p=incubator-singa.git;h=61faa84 ]
SINGA-267 Add spatial mode in batch normalization layer
Added spatial mode in batch normalization layer in C++ implementation,
which corresponds to CUDNN_BATCHNORM_SPATIAL in CuDNN.
Also added logics to automatically detect proper modes in batch
normalization layer, i.e., if input is 2D tensor then batchnorm
layer chooses PER_ACTIVATION mode, if input is 4D tensor then
batchnorm layer chooses SPATIAL mode.
> Add spatial mode in batch normalization layer
> ----------------------------------------------
>
> Key: SINGA-267
> URL: https://issues.apache.org/jira/browse/SINGA-267
> Project: Singa
> Issue Type: Improvement
> Reporter: Wang Ji
>
> Add spatial mode in batch normalization layer in C++ implementation, which corresponds to CUDNN_BATCHNORM_SPATIAL in CuDNN.
> Also add logics to automatically detect proper modes in batch normalization layer, i.e., if input is 2D tensor then batchnorm layer chooses PER_ACTIVATION mode, if input is 4D tensor then batchnorm layer chooses SPATIAL mode.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)