Method to visualize inside of Feature Maps

[Antonio Paes]

Hi guys,

I'm reading a set of paper which show the kernels inside a feature maps.

And I'm doing a search in google, but  I don't find examples about how to implement this.

Someone knows about that?

Thanks!!

[Abhimanyu Dubey]

Load the model in Python.

net = caffe.Net(protofile,modelfile,phase);

For some layer 'convx', you can access the params as net.params['convx'][0] (for weights) and net.params['convx'][1] for biases.

[Antonio Paes]

Thanks Abhimanyu Dubey, but i need of something visual, a region of image after the kernel be applied for example.

[Sudeep Reddy Gaddam]

Even I am looking for what Matthew Zeiler has explained in his talk( https://www.youtube.com/watch?v=ghEmQSxT6tw ). Is any such thing implemented on top of Caffe?? It would be highly useful.

Thanks,
Sudeep.

[Antonio Paes]

This video is good, but is "more of the same", I wanted something implemented with  caffe too. 

[Abhimanyu Dubey]

You can plot the blobs after a forward pass using matplotlib.

[Manuel Rodriguez]

Am Montag, 24. August 2015 08:10:40 UTC+2 schrieb Abhimanyu Dubey:
>  You can plot the blobs after a forward pass using matplotlib.

What he wants is that:
https://www.youtube.com/watch?v=GhqOMJIHD8A (Visualization of CNN. Gesture
Recognition using Convolutional Neural Networks, Ghent University)

Sure, there is a tutorial like
http://nbviewer.ipython.org/github/BVLC/caffe/blob/master/examples/00-classification.ipynb
for visualize the layers of the network, but this tutorial is complicated
and it is not in realtime. Another option is to use GUIs for Caffe like
"Expresso" or "nvidia digits", but installation of these GUIs is complicated.

[Antonio Paes]

Just that Manuel Rodriguez, thank you very much guys.

[Axel Straminsky]

Didn't know about Expresso, thanks Manuel!

[夏志强]

Maybe you have found the method, but you can still have a try on https://github.com/yosinski/deep-visualization-toolbox, which is amazing!

在 2015年8月20日星期四 UTC+8上午9:48:45，Antonio Paes写道：