[
https://issues.apache.org/jira/browse/SINGA-278?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15771781#comment-15771781
]
ASF subversion and git services commented on SINGA-278:
-------------------------------------------------------
Commit bd41fd1d4d3a55bdd8234c357c967a7372a55c83 in incubator-singa's branch
refs/heads/master from wangwei
[ https://git-wip-us.apache.org/repos/asf?p=incubator-singa.git;h=bd41fd1 ]
SINGA-278 Convert trained caffe parameters to singa
Rename the example folder to be 'caffe', which could be used for all caffe
models
Update the readme and instructions to let users type in image paths
Swap the first conv layer's weight matrix for RGB input images
Feed images in RGB order to SINGA
> Convert trained caffe parameters to singa
> -----------------------------------------
>
> Key: SINGA-278
> URL: https://issues.apache.org/jira/browse/SINGA-278
> Project: Singa
> Issue Type: New Feature
> Reporter: Xiangrui
>
> Convert trained parameters of caffe model to singa.
> Run vgg as an example. Some tricks should be noticed:
> 1. The order of image axes in caffe is height, width and channels due to
> opencv implementation, while it is width, height, channels in singa if you
> use python PIL.
> 2. Another problem caused by these two libraries is the order of channels,
> BGR(caffe, opencv) v.s. RGB(singa, PIL).
> 3. It needs to transpose the weight tensor in InnerProduct(Dense) layer.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
