Export Weights for Yosinski Deep Vis Toolbox

[Kelsen Hazelwood]

Hi all,

I currently have a trained pycaffe network, with the snapshot saved. I am wanting to use this model with the Yosinski Deep Visualization Toolbox (https://github.com/yosinski/deep-visualization-toolbox)

In order to do this, you have to edit the settings in a file to point to your network and your network weights. I am new to caffe and currently cannot figure out how to export the network weights in the format that will work for the toolbox. 

# Define critical settings and/or override defaults specified in

# settings.py. Copy this file to settings_local.py in the same

# directory as settings.py and edit. Any settings defined here

# will override those defined in settings.py

# Set this to point to your compiled checkout of caffe

caffevis_caffe_root      = '/Users/kelsen/Documents/caffe'

# Load model: caffenet-yos

# Path to caffe deploy prototxt file. Minibatch size should be 1.

caffevis_deploy_prototxt = '%DVT_ROOT%/models/caffenet-yos/caffenet-yos-deploy.prototxt'

# Path to network weights to load.

caffevis_network_weights = '%DVT_ROOT%/models/caffenet-yos/caffenet-yos-weights'

# Other optional settings; see complete documentation for each in settings.py.

caffevis_data_mean       = '%DVT_ROOT%/models/caffenet-yos/ilsvrc_2012_mean.npy'

caffevis_labels          = '%DVT_ROOT%/models/caffenet-yos/ilsvrc_2012_labels.txt'

caffevis_label_layers    = ('fc8', 'prob')

caffevis_prob_layer      = 'prob'

caffevis_unit_jpg_dir    = '%DVT_ROOT%/models/caffenet-yos/unit_jpg_vis'

caffevis_jpgvis_layers   = ['conv1', 'conv2', 'conv3', 'conv4', 'conv5', 'fc6', 'fc7', 'fc8', 'prob']

caffevis_jpgvis_remap    = {'pool1': 'conv1', 'pool2': 'conv2', 'pool5': 'conv5'}

def caffevis_layer_pretty_name_fn(name):

    return name.replace('pool','p').replace('norm','n')

This is what the settings file looks like. Any advice here would be appreciated. 

Thank you