caffe-unpooling icon indicating copy to clipboard operation
caffe-unpooling copied to clipboard

Published 20 hours ago • verified publisher icon mariolew

DeConv training?

Open deepalianeja opened this issue 8 years ago • 5 comments

I have a trained model and I'm trying your pipeline for unspooling and visualizing the layers. Do I need a new train_val.protxt? Do I need to train again with the DeConv layers? I have updated my deploy.prototxt. Can you please help?

deepalianeja avatar Apr 27 '16 02:04 deepalianeja

My pre-trained model is similar to Alexnet in Caffe.

deepalianeja avatar Apr 27 '16 02:04 deepalianeja

You don't need to train with the DeConv layer, you only need to make the deploy file the similar way as my example.

mariolew avatar Apr 27 '16 02:04 mariolew

@deepalianeja Please refer to examples/deconv_deploy.prototxt and examples/unpooling.ipynb for detail

mariolew avatar Apr 27 '16 02:04 mariolew

Okay. Thanks.

It was looking for conv1t layers and other DeConv laters from the model so I got confused.

Do you also have an example of guided back propagation? I see something but is that complete?

deepalianeja avatar Apr 27 '16 05:04 deepalianeja

@deepalianeja Well, as for the guided back propagation, I think the example just did the right thing but in an ugly way. Do you also think so?

mariolew avatar Apr 27 '16 05:04 mariolew