How to pull weights out of the caffemodel? parsing of caffemodel?

[Farik John]

I want to get the weights of convolution layers from trained caffemodel.

Anyone tried to do this before.

Please guide me.

Thanks.

[Yahia Charif]

Hi John,

do you want to transfer leaning from the pre-trained model ? 

[Clément Fuji Tsang]

use the pycaffe interface, 

just open the model:

import caffe

net = caffe.Net(deploy.prototxt, model.caffemodel, caffe.TEST)

then you can look inside the model like this way:

net.params.items()[layer's number][0 = name of the layer / 1 = layer's learnable data][0 = usually weights and 1 = usually bias]

please notice that you can only see the layer's that can learn (no activation, pooling or dropout).

Clément

[Farik John]

Thanks for your nice answer.

[BIng Frank]

Hi,

Clement, I succesfully  pull weights out of the caffemodel using your method , but when I try to change the weights in caffemodel , I failed . Do you know what the problem is ?
在 2016年5月18日星期三 UTC+8上午12:12:03，Farik John写道：