How to access weights

[Arturo Ghinassi]

If I have something like this:

def my_darknet53(inputs):

      inputs = slim.conv2d(inputs, 32, 3, scope='conv-0')
      inputs = slim.conv2d(inputs, 64, 3, stride=2, scope='conv-1')
      inputs = darknet_block(inputs, 32, scope='dn-block-1')
      inputs = slim.conv2d(inputs, 128, 3, stride=2, scope='conv-3') 
      inputs = slim.repeat(inputs, 2, darknet_block, 64, scope='dn-block-2')
      inputs = slim.conv2d(inputs, 256, 3, stride=2, scope='conv-4')
      inputs = slim.repeat(inputs, 8, darknet_block, 128, scope='dn-block-3')
      inputs = slim.conv2d(inputs, 512, 3, stride=2, scope='conv-5')
      inputs = slim.repeat(inputs, 8, darknet_block, 256, scope='dn-block-4')
      inputs = slim.conv2d(inputs, 1024, 3, stride=2, scope='conv-6')
      inputs = slim.repeat(inputs, 4, darknet_block, 512, scope='dn-block-5')
      
      return inputs

How can I access the weights of 'conv-0' for example?

[Mark Sandler]

You can get get all your variables using somethng like this: https://stackoverflow.com/questions/36533723/tensorflow-get-all-variables-in-scope

…

On Tue, Jan 17, 2023 at 3:53 AM Arturo Ghinassi ***@***.***> wrote: If I have something like this: def my_darknet53(inputs): inputs = slim.conv2d(inputs, 32, 3, scope='conv-0') inputs = slim.conv2d(inputs, 64, 3, stride=2, scope='conv-1') inputs = darknet_block(inputs, 32, scope='dn-block-1') inputs = slim.conv2d(inputs, 128, 3, stride=2, scope='conv-3') inputs = slim.repeat(inputs, 2, darknet_block, 64, scope='dn-block-2') inputs = slim.conv2d(inputs, 256, 3, stride=2, scope='conv-4') inputs = slim.repeat(inputs, 8, darknet_block, 128, scope='dn-block-3') inputs = slim.conv2d(inputs, 512, 3, stride=2, scope='conv-5') inputs = slim.repeat(inputs, 8, darknet_block, 256, scope='dn-block-4') inputs = slim.conv2d(inputs, 1024, 3, stride=2, scope='conv-6') inputs = slim.repeat(inputs, 4, darknet_block, 512, scope='dn-block-5') return inputs How can I access the weights of 'conv-0' for example? — Reply to this email directly, view it on GitHub <#22>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABVIDWEMETDS73TKKREWBE3WS2B2JANCNFSM6AAAAAAT5YPASE> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>