def f(x):
coloring_group, m = get_m_group(x, coloring_m, axis)
if coloring_group > 1:
splits = Split(coloring_group, axis)(x)
cs = []
us = []
for i, split in enumerate(splits):
split_c = factor_conv_layer(
number_of_classes=number_of_classes,
name=name + '_c' + str(i),
filters=m,
filters_emb=filters_emb,
use_bias=False)([split, cls])
split_u = uncoditional_conv_layer(kernel_size=(1, 1),
filters=m,
name=name + '_u' +
str(i))(split)
cs.append(split_c)
us.append(split_u)
c = tf.keras.layers.Concatenate(axis)(cs)
u = tf.keras.layers.Concatenate(axis)(us)
else:
c = factor_conv_layer(number_of_classes=number_of_classes,
name=name + '_c',
filters=K.int_shape(x)[axis],
filters_emb=filters_emb,
use_bias=False)([x, cls])
u = uncoditional_conv_layer(kernel_size=(1, 1),
filters=K.int_shape(x)[axis],
name=name + '_u')(x)
out = Add(name=name + '_a')([c, u])
return out
def f(x):
coloring_group, m = get_m_group(x, coloring_m, axis)
if coloring_group > 1:
splits = Split(coloring_group, axis)(x)
outs = []
for i, split in enumerate(splits):
split_out = uncoditional_conv_layer(filters=m,
kernel_size=(1, 1),
name=name +
str(i))(split)
outs.append(split_out)
out = tf.keras.layers.Concatenate(axis)(outs)
else:
out = uncoditional_conv_layer(filters=K.int_shape(x)[axis],
kernel_size=(1, 1),
name=name)(x)
return out
def f(x):
coloring_group, m = get_m_group(x, coloring_m, axis)
if coloring_group > 1:
splits = Split(coloring_group, axis)(x)
outs = []
for i, split in enumerate(splits):
split_out = conditional_conv_layer(
filters=m,
number_of_classes=number_of_classes,
name=name + str(i))([split, cls])
outs.append(split_out)
out = tf.keras.layers.Concatenate(axis)(outs)
else:
out = conditional_conv_layer(
filters=K.int_shape(x)[axis],
number_of_classes=number_of_classes,
name=name)([x, cls])
return out
def f(x):
coloring_group, m = get_m_group(x, coloring_m, axis)
c = ConditionalCenterScale(number_of_classes=number_of_classes,
axis=axis,
name=name + '_c')([x, cls])
if coloring_group > 1:
splits = Split(coloring_group, axis)(x)
us = []
for i, split in enumerate(splits):
split_u = uncoditional_conv_layer(kernel_size=(1, 1),
filters=m,
name=name + '_u' +
str(i))(split)
us.append(split_u)
u = tf.keras.layers.Concatenate(axis)(us)
else:
u = uncoditional_conv_layer(kernel_size=(1, 1),
filters=K.int_shape(x)[axis],
name=name + '_u')(x)
out = Add(name=name + '_a')([c, u])
return out