call forward_prefilled on only one layer

[Amir Abdi]

I'm wondering if there is a way to "forward" the network on only one layer.

If there is no way to do that, how would it benefit us to set the data in middle blobs to something else as they will get replaced by new data when the whole network runs forward!

I want to load data in a bottom layer, run the forward on that layer, and see the outputs of that layer.

[Amir Abdi]

I believe there is a function available in pycaffe, but nothing is implemented for matcaffe.
I need to do this in MATLAB.

On Thursday, March 3, 2016 at 5:13:31 PM UTC-8, Amir Abdi wrote:

I believe there is a function available in pycaffe, but nothing is implemented for matcaffe.

On Thursday, March 3, 2016 at 5:07:42 PM UTC-8, Amir Abdi wrote:

I'm using the  MATLAB interface.

On Thursday, March 3, 2016 at 5:02:30 PM UTC-8, Amir Abdi wrote:

For now, what I'm doing is I define a separate network for each layer! So that each network only holds one single layer and I can call forward() on them.

[Jan C Peters]

There definitely is the possibility of doing that in pycaffe, with net.forward(from=..., to=...).

I just had a brief look at the matcaffe code, and it seems this is indeed impossible there. So you have two options: switch to python, or adjust the matcaffe API yourself to expose this functionality to MATLAB. Maybe you're lucky and somebody already did that. Search the PRs and forks.

Jan