As much as I know in keras border_mode
is used for padding. What I need to know how can I specify a value like Padding/border_mode=2
in keras? The problem is border_mode
take only two values: valid
and same
So any help?
p.s already asked on data science. But may be it is beta so getting no response :(
You can try a ZeroPadding2D layer:
https://keras.io/layers/convolutional/#zeropadding2d
Note: the Keras Convolution2D layer is backed by either the theano.nnet.conv2d or tf.nn.conv2d. The latter doesn't seem to support custom padding (although there's tf.pad for this).