guruucsd / cnn_visualization

An implementation of CNN visualization.

Convolutional Neural Net visualization

This is our implementation (plus a demo) of CNN feature visualization (Zeiler & Fergus, 2013).

The way to just look at the ipython notebook. Use link: http://nbviewer.ipython.org/github/guruucsd/CNN_visualization/blob/master/urban_tribe/caffe/deconv_demo.ipynb

The visualization result is not as clear as the paper shows. We assume it's because they use different architecture and we have a small dataset for visualization. However, it may due to undetected bugs (but we hope not..) If you find any mistakes in this code, please tell us!

References

Zeiler, M. D. and Fergus, R. (2013) Visualizing and Understanding Convolutional Networks. http://arxiv.org/abs/1311.2901

To run

This notebook should be run from the guru2.ucsd.edu server. Installation of caffe can be quite challenging, and this notebook depends on a particular urban_tribes model that only exists on that server.

To run,

• ssh -X [username@]guru2.ucsd.edu
• Clone this repository
• Run ipython notebook and open the CNN_visualization/urban_tribe/caffe/deconv_demo.ipynb file from the browser.

cnn_visualization's Issues

regarding .pkl file used

Hi all,
I went through this code, and found this pickle file used in the following path :
files = '/home/feiyu1990/local/urban_tribe/caffe/CNN_cross1/dataset_alignment_5/0.0.pkl'
can anyone please explain me what this file contain and why it has been used. Thanks in advance.

Notebook page shows Error 503 No healthy backends

The ipynb link in the Readme shows Error 503 No healthy backends.

Error 503 No healthy backends
No healthy backends
Guru Mediation:
Details: cache-ams4133-AMS 1513241057 2271580453
Varnish cache server

Can I run the code?

hi！I am new here.And also new for CNN_visualization……I have too many questions……
Like, How can I run the code? Should I have to have a 'username' in [username@]guru2.ucsd.edu?
Thanks!And I really hope that you can help me solve my problem at your convenience……Thank you very much!