def layer4_1(x):
pad4_1 = F.pad(x, (1, 1, 1, 1), 'reflect')
conv4_1 = PF.convolution(
pad4_1, 256, kernel=(
3, 3), stride=(
1, 1), name='layer4_1.1')
conv4_1 = F.instance_normalization(
conv4_1, gamma=None, beta=None, channel_axis=1)
conv4_1 = PF.prelu(conv4_1, name='layer4_1.3')
pad4_2 = F.pad(conv4_1, (1, 1, 1, 1), 'reflect')
conv4_2 = PF.convolution(
pad4_2, 64, kernel=(
3, 3), stride=(
1, 1), name='layer4_1.5')
conv4_2 = F.instance_normalization(
conv4_2, gamma=None, beta=None, channel_axis=1)
conv4_2 = PF.prelu(conv4_2, name='layer4_1.7')
up4_1 = F.interpolate(
conv4_2,
scale=(
2,
2),
mode='nearest',
align_corners=False)
return up4_1
def layer3_1(x):
pad3_1 = F.pad(x, (1, 1, 1, 1), 'reflect')
conv3_1 = PF.convolution(
pad3_1, 128, kernel=(
3, 3), stride=(
1, 1), name='layer3_1.1')
conv3_1 = F.instance_normalization(
conv3_1, gamma=None, beta=None, channel_axis=1)
conv3_1 = PF.prelu(conv3_1, name='layer3_1.3')
pad3_2 = F.pad(conv3_1, (1, 1, 1, 1), 'reflect')
conv3_2 = PF.convolution(
pad3_2, 64, kernel=(
3, 3), stride=(
1, 1), name='layer3_1.5')
conv3_2 = F.instance_normalization(
conv3_2, gamma=None, beta=None, channel_axis=1)
conv3_2 = PF.prelu(conv3_2, name='layer3_1.7')
return conv3_2
def res_block(x, out_ch, name):
with nn.parameter_scope(name):
residual = x
out = F.pad(x, (1, 1, 1, 1), 'reflect')
out = PF.convolution(
out, out_ch, kernel=(
3, 3), stride=(
1, 1), name='conv1')
out = F.instance_normalization(
out, gamma=None, beta=None, channel_axis=1)
out = PF.prelu(out)
out = F.pad(out, (1, 1, 1, 1), 'reflect')
out = PF.convolution(
out, out_ch, kernel=(
3, 3), stride=(
1, 1), name='conv2')
out = F.instance_normalization(
out, gamma=None, beta=None, channel_axis=1)
out += residual
out = PF.prelu(out)
return out
def test_pf_prelu_execution(g_rng, inshape, base_axis, shared, slope_init,
fix_parameters):
slope_shape = tuple() if shared else (inshape[base_axis], )
slope_init = process_param_init(slope_init, slope_shape, g_rng)
kw = {}
insert_if_not_none(kw, 'slope_init', slope_init)
insert_if_not_default(kw, 'base_axis', base_axis, 1)
insert_if_not_default(kw, 'shared', shared, True)
insert_if_not_default(kw, 'fix_parameters', fix_parameters, False)
x = nn.Variable.from_numpy_array(g_rng.randn(*inshape))
# Check execution
y = PF.prelu(x, **kw)
y.forward()
y.backward()
# Check values
# TODO
# Check args
assert y.parent.info.type_name == 'PReLU'
args = y.parent.info.args
assert args['base_axis'] == base_axis
# Check created parameters
assert y.parent.inputs[0] == x
assert len(y.parent.inputs) == 2
assert len(nn.get_parameters()) == 1
slope = nn.get_parameters()['prelu/slope']
assert slope.shape == slope_shape
assert slope.need_grad
assert y.parent.inputs[1].need_grad == (not fix_parameters)
if isinstance(slope_init, np.ndarray):
assert np.allclose(slope_init, slope.d)
def af(x):
return PF.prelu(x)
def af(x, name=None):
return PF.prelu(x, name=name)
def network(self, x, test=False):
# Input:x -> 3,256,256
# Convolution_5 -> 16,256,256
h = PF.convolution(x, 16, (3, 3), (1, 1), name='Convolution_5')
# BatchNormalization_9
h = PF.batch_normalization(h, (1, ),
0.9,
0.0001,
not test,
name='BatchNormalization_9')
# PReLU_8
h = PF.prelu(h, 1, False, name='PReLU_8')
# Convolution_6
h = PF.convolution(h, 16, (3, 3), (1, 1), name='Convolution_6')
# BatchNormalization_5
h = PF.batch_normalization(h, (1, ),
0.9,
0.0001,
not test,
name='BatchNormalization_5')
# PReLU_7
h = PF.prelu(h, 1, False, name='PReLU_7')
# Convolution_4
h = PF.convolution(h, 16, (3, 3), (1, 1), name='Convolution_4')
# BatchNormalization_2
h = PF.batch_normalization(h, (1, ),
0.9,
0.0001,
not test,
name='BatchNormalization_2')
# PReLU_6
h = PF.prelu(h, 1, False, name='PReLU_6')
# MaxPooling_2 -> 16,128,128
h = F.max_pooling(h, (2, 2), (2, 2), False)
# Convolution_2 -> 32,128,128
h = PF.convolution(h, 32, (3, 3), (1, 1), name='Convolution_2')
# BatchNormalization_4
h = PF.batch_normalization(h, (1, ),
0.9,
0.0001,
not test,
name='BatchNormalization_4')
# PReLU_5
h = PF.prelu(h, 1, False, name='PReLU_5')
# MaxPooling -> 32,64,64
h = F.max_pooling(h, (2, 2), (2, 2), False)
# Convolution_3 -> 64,64,64
h = PF.convolution(h, 64, (3, 3), (1, 1), name='Convolution_3')
# BatchNormalization
h = PF.batch_normalization(h, (1, ),
0.9,
0.0001,
not test,
name='BatchNormalization')
# PReLU_4
h = PF.prelu(h, 1, False, name='PReLU_4')
# MaxPooling_4 -> 64,32,32
h = F.max_pooling(h, (2, 2), (2, 2), False)
# Convolution_7 -> 128,32,32
h = PF.convolution(h, 128, (3, 3), (1, 1), name='Convolution_7')
# BatchNormalization_7
h = PF.batch_normalization(h, (1, ),
0.9,
0.0001,
not test,
name='BatchNormalization_7')
# PReLU_3
h = PF.prelu(h, 1, False, name='PReLU_3')
# MaxPooling_3 -> 128,16,16
h = F.max_pooling(h, (2, 2), (2, 2), False)
# Convolution_8 -> 256,16,16
h = PF.convolution(h, 256, (3, 3), (1, 1), name='Convolution_8')
# BatchNormalization_10
h = PF.batch_normalization(h, (1, ),
0.9,
0.0001,
not test,
name='BatchNormalization_10')
# PReLU_2
h = PF.prelu(h, 1, False, name='PReLU_2')
# MaxPooling_5 -> 256,8,8
h = F.max_pooling(h, (2, 2), (2, 2), False)
# Convolution -> 512,8,8
h = PF.convolution(h, 512, (3, 3), (1, 1), name='Convolution')
# BatchNormalization_8
h = PF.batch_normalization(h, (1, ),
0.9,
0.0001,
not test,
name='BatchNormalization_8')
# PReLU
h = PF.prelu(h, 1, False, name='PReLU')
# AveragePooling -> 512,1,1
h = F.average_pooling(h, (8, 8), (8, 8))
# BatchNormalization_6
h = PF.batch_normalization(h, (1, ),
0.9,
0.0001,
not test,
name='BatchNormalization_6')
# PReLU_9
h = PF.prelu(h, 1, False, name='PReLU_9')
# Affine -> 1
h = PF.affine(h, (1, ), name='Affine')
# BatchNormalization_3
h = PF.batch_normalization(h, (1, ),
0.9,
0.0001,
not test,
name='BatchNormalization_3')
# y'
h = F.sigmoid(h)
return h
