Dear Authors,
I have several questions about the minibatch_std_layer function.
1.original code:
# Calculate the standard deviation over the group [M,W,H,C]minibatch = K.backend.square(minibatch + 1e8)
Since here standard deviation should be calculated, I think this line should be
minibatch = K.backend.sqrt(minibatch + 1e-8)
original code:
# Take average over feature maps and pixels [M,1,1,1]minibatch = tf.reduce_mean(minibatch, axis=[1,2,4], keepdims=True)
Why is axis=[1,2,4]?
I think it should be axis=[1,2,3].
original code:
shape = list(K.int_shape(input))
Does "input" mean "layer"?
I do not understand why group_size is considered here.
Can you please explain why group_size is necessary?
Dear Authors, I have several questions about the minibatch_std_layer function. 1.original code:
# Calculate the standard deviation over the group [M,W,H,C]minibatch = K.backend.square(minibatch + 1e8)Since here standard deviation should be calculated, I think this line should beminibatch = K.backend.sqrt(minibatch + 1e-8)original code:
# Take average over feature maps and pixels [M,1,1,1]minibatch = tf.reduce_mean(minibatch, axis=[1,2,4], keepdims=True)Why isaxis=[1,2,4]? I think it should beaxis=[1,2,3].original code:
shape = list(K.int_shape(input))Does "input" mean "layer"?I do not understand why group_size is considered here. Can you please explain why group_size is necessary?