Does the caffe has the plan to implement locally connected unshared weight layer?

[gang zhang]

recently, i want to implement a face recognition demo.there is locally connected unshared weight layer in the demo net,but caffe is not support.

I try to implement to the layer,but find it is a  difficult job,so i want to konw  the  caffe whether or not has the plan to implement it. 

[Yangqing Jia]

If memory is not an issue - stack an im2col layer, a fully connected layer, and a reshape layer will sort of solve your problem.

Yangqing

On Thu, Sep 25, 2014 at 4:04 AM, gang zhang <xtmq...@gmail.com> wrote:

recently, i want to implement a face recognition demo.there is locally connected unshared weight layer in the demo net,but caffe is not support.

I try to implement to the layer,but find it is a  difficult job,so i want to konw  the  caffe whether or not has the plan to implement it. 

--
You received this message because you are subscribed to the Google Groups "Caffe Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to caffe-users...@googlegroups.com.
To post to this group, send email to caffe...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/caffe-users/c1a54ccc-7a78-4edd-99d8-b6959dd86087%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ran Manor]

Hi Yangqing,

Can you please give a more concrete examples on how to do this?

How can I define with the im2col layer how to split the data per convolution filter?

Thanks,

Ran

[Amir Alush]

This is a good implementation, I've tested it and it seems to converge nicely.

https://github.com/BVLC/caffe/pull/1271

[Ran Manor]

Thanks. This looks good, but how do I define the local regions in this layer?

[Kevin W Li]

Sorry for bringing up the old discussion. Just wondering if there is any docs on im2col layer? I would like to implement this before the ongoing locally-connected layer gets merged. Not an experienced programmer so any help is greatly appreciated. Thank you.