How can I visualize the filters from the conv2d function ?

[VigneshSrinivasan10]

In tensorflow, it is possible to visualize the filters using the image summary. How can I do the same in prettytensor ?

Edit: To rephrase the question, I wanted to access the 'weights' form the first convolution layer. So if the name of my first conv layer was conv1 as in conv2d(9, 64, stride=1, init=tf.truncated_normal_initializer(stddev=0.001),edges='SAME',name='conv1')
To access the variable - conv1/weights, I used

with tf.variable_scope('conv1') as scope:
        tf.get_variable_scope().reuse_variables()
        weights = tf.get_variable('weights')

But got an error: ValueError: Variable conv1/weights does not exist, disallowed

What am I doing wrong here. Please suggest the correct way to extract the weights!

Thanks in advance!

[eiderman]

Are you in the same scope where you created the variable? variable_scopes are nested.

The following code properly captures the weights:

with tf.Graph().as_default() as g:
  x = pt.wrap(np.zeros((4, 10, 10, 4), dtype=np.float32)).conv2d(9, 64, stride=1, weights=tf.truncated_normal_initializer(stddev=0.001),edges='SAME',name='conv1')

  with tf.variable_scope('conv1') as scope:
      tf.get_variable_scope().reuse_variables()
      weights = tf.get_variable('weights')
      print weights

To see the full variable names, you can run this: print [v.name for v in tf.all_variables()].

You can also use: x.layer_parameters['weights'] if you have a handle on the layer to get the weights used in the graph.

PS I personally find the style of with tf.variable_scope('conv1', reuse=True) to be a little easier to read.

[VigneshSrinivasan10]

print [v.name for v in tf.all_variables()] got me all the variable names and then I was able to extract the weights I wanted. Thank you!