def get_model(img_shape=(128, 128, 3), filter_num=16):
inputs = keras.layers.Input(shape=(img_shape))
o = inputs
for l, s in reversed(layer_dict.items()):
if l == img_shape[0]:
o = keras.layers.Conv2D(filters=filter_num * s,
kernel_size=3,
strides=1,
padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
if l < img_shape[0]:
o = keras.layers.Conv2D(filters=filter_num * s,
kernel_size=3,
strides=2,
padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
for l, s in layer_dict.items():
if l == img_shape[0] // 2:
break
o = keras.layers.Conv2DTranspose(filters=filter_num * s,
kernel_size=3,
strides=2,
padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
o = keras.layers.Conv2DTranspose(filters=256,
kernel_size=3,
strides=2,
padding="SAME")(o)
return keras.Model(inputs=inputs, outputs=o)
def build(self, input_shape):
noise = keras.layers.Input(shape=(self.latent_dim, ))
for i, f in layer_dict.items():
if i == min(layer_dict.keys()):
o = keras.layers.Dense(i * i * f * self.filter_num)(noise)
o = keras.layers.Reshape((i, i, f * self.filter_num))(o)
else:
o = keras.layers.Conv2DTranspose(
filters=f * self.filter_num,
kernel_size=5,
strides=2,
padding="SAME",
kernel_initializer=keras.initializers.RandomNormal(
mean=0.0, stddev=0.02))(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
if i == input_shape[0] // 2:
break
o = keras.layers.Conv2DTranspose(
filters=input_shape[-1],
kernel_size=5,
strides=2,
padding='SAME',
kernel_initializer=keras.initializers.RandomNormal(mean=0.0,
stddev=0.02),
activation="tanh")(o)
self.gen = keras.Model(inputs=noise, outputs=o)
img = keras.layers.Input(shape=input_shape)
o = AugmentLayer()(img)
for i, f in reversed(layer_dict.items()):
if i < input_shape[0]:
o = keras.layers.Conv2D(
filters=f * self.filter_num,
kernel_size=5,
strides=2,
padding="SAME",
kernel_initializer=keras.initializers.RandomNormal(
mean=0.0, stddev=0.02))(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
o = keras.layers.Flatten()(o)
o = keras.layers.Dense(1)(o)
self.dis = keras.Model(inputs=img, outputs=o)
def build(self, input_shape):
noise = keras.layers.Input(shape=(self.latent_dim,))
for i, f in layer_dict.items():
if i == min(layer_dict.keys()):
o = noise[:, tf.newaxis, tf.newaxis, :]
o = convt(f * self.filter_num, kernel_size=min(layer_dict.keys()),
strides=1, padding="VALID")(o)
o = norm_layer(o)
o = act_layer(o)
o = conv(f * self.filter_num, kernel_size=5, strides=1)(o)
o = norm_layer(o)
o = act_layer(o)
else:
o = convt(f * self.filter_num, kernel_size=3, strides=2)(o)
o = norm_layer(o)
o = act_layer(o)
o = conv(f * self.filter_num, kernel_size=5, strides=1)(o)
o = norm_layer(o)
o = act_layer(o)
if i == input_shape[0]:
break
o = conv(input_shape[-1], kernel_size=1, strides=1)(o)
o = keras.layers.Activation("tanh")(o)
self.gen = keras.Model(inputs=noise, outputs=o)
img = keras.layers.Input(shape=input_shape)
o = AugmentLayer()(img)
o = img
os = [o]
for i, f in reversed(layer_dict.items()):
if i < input_shape[0]:
o = conv(f*self.filter_num, kernel_size=3, strides=2)(o)
o = norm_layer(o)
o = act_layer(o)
o = conv(f * self.filter_num, kernel_size=5, strides=1)(o)
o = norm_layer(o)
o = act_layer(o)
if i >= 32:
o_ = conv(input_shape[-1], kernel_size=1, strides=1)(o)
o_ = norm_layer(o_)
o_ = keras.layers.Activation("tanh")(o_)
os.append(o_)
o = keras.layers.Flatten()(o)
self.dis = keras.Model(inputs=img, outputs=[o]+os)
def get_model(img_shape=(128, 128, 3), filter_num=16):
inputs = keras.layers.Input(shape=(img_shape))
o = inputs
o_dict = dict()
for l, s in reversed(layer_dict.items()):
if l == img_shape[0]:
o = keras.layers.Conv2D(filters=filter_num * s, kernel_size=3, strides=1, padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
o_dict[l] = o
if l < img_shape[0]:
o = keras.layers.Conv2D(filters=filter_num * s, kernel_size=3, strides=2, padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
o_dict[l] = o
for l, s in layer_dict.items():
if l == img_shape[0]:
break
o = keras.layers.Conv2DTranspose(filters=filter_num * layer_dict[l * 2], kernel_size=3, strides=2, padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
o_ = keras.layers.Conv2D(filters=filter_num * layer_dict[l * 2], kernel_size=1, strides=1, padding="SAME")(o_dict[l * 2])
o_ = keras.layers.BatchNormalization()(o_)
o_ = keras.layers.LeakyReLU(0.2)(o_)
o = keras.layers.Add()([o, o_])
o_dict[l * 2] = o
os = []
for l, o in o_dict.items():
o_ = keras.layers.Conv2DTranspose(filters=filter_num * layer_dict[img_shape[0]], kernel_size=1, strides=img_shape[0]//l, padding="SAME")(o)
o_ = keras.layers.BatchNormalization()(o_)
o_ = keras.layers.LeakyReLU(0.2)(o_)
os.append(o_)
o = keras.layers.Concatenate(axis=-1)(os)
o = keras.layers.Conv2D(filters=256, kernel_size=1, strides=1, padding="SAME")(o)
return keras.Model(inputs=inputs, outputs=o)
def build(self, input_shape):
super(SpectralGAN, self).build(input_shape)
img = keras.layers.Input(shape=input_shape)
o = img
for i, f in reversed(layer_dict.items()):
if i < self.img_shape[0]:
o = tfa.layers.SpectralNormalization(
keras.layers.Conv2D(
filters=f * self.filter_num,
kernel_size=5,
strides=2,
padding="SAME",
kernel_initializer=keras.initializers.RandomNormal(
mean=0.0, stddev=0.02)))(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
o = keras.layers.GlobalAveragePooling2D()(o)
o = keras.layers.Dense(1)(o)
def get_model(img_shape=(128, 128, 3), filter_num=16):
inputs = keras.layers.Input(shape=(img_shape))
o = inputs
m = keras.applications.ResNet50(include_top=False, weights=None, input_shape=img_shape)
o = m(o)
org_o = keras.layers.Conv2D(filters=filter_num * layer_dict[o.shape[1]], kernel_size=1, strides=1)(o)
org_o = keras.layers.BatchNormalization()(org_o)
org_o = keras.layers.LeakyReLU(0.2)(org_o)
rev_o = keras.layers.Conv2D(filters=filter_num * layer_dict[o.shape[1]], kernel_size=1, strides=1)(o)
rev_o = keras.layers.BatchNormalization()(rev_o)
rev_o = keras.layers.LeakyReLU(0.2)(rev_o)
att = keras.layers.Activation("sigmoid")(-org_o)
att_o = keras.layers.Multiply()([att, rev_o])
att_o = keras.layers.Subtract()([org_o, att_o])
rev_o = -rev_o
for l, s in layer_dict.items():
if l == img_shape[0]:
break
if l >= o.shape[1]:
org_o = keras.layers.Conv2DTranspose(filters=filter_num * layer_dict[l * 2], kernel_size=3, strides=2, padding="SAME")(org_o)
org_o = keras.layers.BatchNormalization()(org_o)
org_o = keras.layers.LeakyReLU(0.2)(org_o)
rev_o = keras.layers.Conv2DTranspose(filters=filter_num * layer_dict[l * 2], kernel_size=3, strides=2, padding="SAME")(rev_o)
rev_o = keras.layers.BatchNormalization()(rev_o)
rev_o = keras.layers.LeakyReLU(0.2)(rev_o)
att_o = keras.layers.Conv2DTranspose(filters=filter_num * layer_dict[l * 2], kernel_size=3, strides=2, padding="SAME")(att_o)
att_o = keras.layers.BatchNormalization()(att_o)
att_o = keras.layers.LeakyReLU(0.2)(att_o)
o = keras.layers.Concatenate(axis=-1)([org_o, rev_o, att_o])
o = keras.layers.Conv2DTranspose(filters=256, kernel_size=1, strides=1, padding="SAME")(o)
return keras.Model(inputs=inputs, outputs=o)
def get_model(img_shape=(128, 128, 3), filter_num=16):
inputs = keras.layers.Input(shape=(img_shape))
o = inputs
for l, s in reversed(layer_dict.items()):
if l == img_shape[0]:
o = keras.layers.Conv2D(filters=filter_num * s,
kernel_size=3,
strides=1,
padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
if l < img_shape[0]:
o = keras.layers.Conv2D(filters=filter_num * s,
kernel_size=3,
strides=2,
padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
o = keras.layers.Conv2D(filters=filter_num * layer_dict[o.shape[1]],
kernel_size=3,
strides=1,
dilation_rate=2,
padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
pool = keras.layers.AveragePooling2D(pool_size=img_shape[0] //
o.shape[1])(inputs)
c1 = keras.layers.Conv2D(filters=filter_num * layer_dict[o.shape[1]],
kernel_size=1,
strides=1,
padding="SAME")(o)
c1 = keras.layers.BatchNormalization()(c1)
c1 = keras.layers.LeakyReLU(0.2)(c1)
c6 = keras.layers.Conv2D(filters=filter_num * layer_dict[o.shape[1]],
kernel_size=3,
strides=1,
dilation_rate=6,
padding="SAME")(o)
c6 = keras.layers.BatchNormalization()(c6)
c6 = keras.layers.LeakyReLU(0.2)(c6)
c12 = keras.layers.Conv2D(filters=filter_num * layer_dict[o.shape[1]],
kernel_size=3,
strides=1,
dilation_rate=12,
padding="SAME")(o)
c12 = keras.layers.BatchNormalization()(c12)
c12 = keras.layers.LeakyReLU(0.2)(c12)
c18 = keras.layers.Conv2D(filters=filter_num * layer_dict[o.shape[1]],
kernel_size=3,
strides=1,
dilation_rate=18,
padding="SAME")(o)
c18 = keras.layers.BatchNormalization()(c18)
c18 = keras.layers.LeakyReLU(0.2)(c18)
o = keras.layers.Concatenate(axis=-1)([pool, c1, c6, c12, c18])
o = keras.layers.Conv2D(filters=filter_num * layer_dict[o.shape[1]],
kernel_size=1,
strides=1,
padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
for l, s in layer_dict.items():
if l == img_shape[0] // 2:
break
o = keras.layers.Conv2DTranspose(filters=filter_num *
layer_dict[o.shape[1]],
kernel_size=1,
strides=2,
padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
o = keras.layers.Conv2DTranspose(filters=256,
kernel_size=1,
strides=2,
padding="SAME")(o)
return keras.Model(inputs=inputs, outputs=o)
def build(self, input_shape):
self.M = keras.Sequential(
[keras.layers.InputLayer(input_shape=(self.latent_dim, ))])
for _ in range(8):
self.M.add(
keras.layers.Dense(self.latent_dim,
kernel_initializer="he_normal"))
self.M.add(keras.layers.LeakyReLU(0.2))
style_list = []
inp = keras.layers.Input(shape=(self.latent_dim, ))
noise = keras.layers.Input(shape=input_shape)
outs = []
self.n = 0
for i, f in layer_dict.items():
if i == min(layer_dict.keys()):
style = keras.layers.Input(shape=(self.latent_dim, ))
style_list.append(style)
self.n += 1
o = keras.layers.Dense(self.latent_dim)(inp)
o = o[:, tf.newaxis, tf.newaxis, :]
o = up(o, size=i)
o = Conv2DMod(f * self.filter_num, kernel_size=3,
strides=1)([o, style])
noise_crop = keras.layers.Cropping2D(
(input_shape[0] - i) // 2)(noise)
o_n = conv(f * self.filter_num, kernel_size=1,
strides=1)(noise_crop)
o = keras.layers.Add()([o, o_n])
o = act_layer(o)
o = Conv2DMod(f * self.filter_num, kernel_size=3,
strides=1)([o, style])
o_n = conv(f * self.filter_num, kernel_size=1,
strides=1)(noise_crop)
o = keras.layers.Add()([o, o_n])
o = act_layer(o)
rgb = Conv2DMod(input_shape[-1],
kernel_size=1,
strides=1,
demod=False)([o, style])
outs.append(up(rgb, size=input_shape[0] // i))
else:
style = keras.layers.Input(shape=(self.latent_dim, ))
style_list.append(style)
self.n += 1
o = up(o, size=2)
o = Conv2DMod(f * self.filter_num, kernel_size=3,
strides=1)([o, style])
noise_crop = keras.layers.Cropping2D(
(input_shape[0] - i) // 2)(noise)
o_n = conv(f * self.filter_num, kernel_size=1,
strides=1)(noise_crop)
o = keras.layers.Add()([o, o_n])
o = act_layer(o)
o = Conv2DMod(f * self.filter_num, kernel_size=3,
strides=1)([o, style])
o_n = conv(f * self.filter_num, kernel_size=1,
strides=1)(noise_crop)
o = keras.layers.Add()([o, o_n])
o = act_layer(o)
rgb = Conv2DMod(input_shape[-1],
kernel_size=1,
strides=1,
demod=False)([o, style])
outs.append(up(rgb, size=input_shape[0] // i))
if i == input_shape[0]:
self.m = self.n
break
o = keras.layers.Add()(outs)
# o = keras.layers.Activation("tanh")(o)
self._gen = keras.Model(inputs=[inp, noise] + style_list, outputs=o)
style_list_n = []
o_list = []
inp = keras.layers.Input(shape=(self.latent_dim, ))
noise = keras.layers.Input(shape=input_shape)
for _ in style_list:
style = keras.layers.Input(shape=(self.latent_dim, ))
style_list_n.append(style)
o_list.append(self.M(style))
o = self._gen([inp, noise] + o_list)
self.gen = keras.Model(inputs=[inp, noise] + style_list_n, outputs=o)
img = keras.layers.Input(shape=input_shape)
o = AugmentLayer()(img)
for i, f in reversed(layer_dict.items()):
if i < input_shape[0]:
res = conv(f * self.filter_num, kernel_size=3, strides=2)(o)
o = conv(f * self.filter_num, kernel_size=3, strides=2)(o)
o = norm_layer(o)
o = act_layer(o)
o = conv(f * self.filter_num, kernel_size=3, strides=1)(o)
o = norm_layer(o)
o = keras.layers.Add()([res, o])
o = act_layer(o)
o = keras.layers.Flatten()(o)
self.dis = keras.Model(inputs=img, outputs=o)
self.perform_gp = True
self.perform_pl = True
self.pl_mean = 0
self.pl_length = 0.
def get_model(img_shape=(128, 128, 3), filter_num=16, version="v1"):
inputs = keras.layers.Input(shape=(img_shape))
o = inputs
o_dict = dict()
l_list = []
for l, s in reversed(layer_dict.items()):
if l == img_shape[0]:
o = keras.layers.Conv2D(filters=filter_num * s, kernel_size=3, strides=1, padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
o_dict[l] = o
l_list.append(l)
if l < img_shape[0]:
l_list.append(l)
t_dict = dict()
for l in l_list[:-1]:
for l_ in l_list:
t_dict.setdefault(l_, [])
if l == l_:
o = keras.layers.Conv2D(filters=filter_num * layer_dict[l_], kernel_size=3, strides=1,
padding="SAME")(o_dict[l])
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
t_dict[l_].append(o)
elif l > l_:
o = keras.layers.Conv2D(filters=filter_num * layer_dict[l_], kernel_size=3, strides=l // l_,
padding="SAME")(o_dict[l])
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
t_dict[l_].append(o)
else:
o = keras.layers.Conv2DTranspose(filters=filter_num * layer_dict[l_], kernel_size=3,
strides=l_ // l, padding="SAME")(o_dict[l])
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
t_dict[l_].append(o)
for l, os in t_dict.items():
if len(os) == 1:
o = os[0]
else:
o = keras.layers.Concatenate(axis=-1)(os)
o = keras.layers.Conv2D(filters=filter_num * layer_dict[l], kernel_size=3, strides=1, padding="SAME")(o)
o = keras.layers.BatchNormalization()(o)
o = keras.layers.LeakyReLU(0.2)(o)
o_dict[l] = o
if version == "v1":
o = keras.layers.Conv2D(256, kernel_size=3, strides=1, padding="SAME")(o_dict[img_shape[0]])
elif version == "v2":
os = []
for l, o in o_dict.items():
o_ = keras.layers.Conv2DTranspose(filters=filter_num * layer_dict[l], kernel_size=3, strides=img_shape[0]//l, padding="SAME")(o)
o_ = keras.layers.BatchNormalization()(o_)
o_ = keras.layers.LeakyReLU(0.2)(o_)
os.append(o_)
o = keras.layers.Concatenate(axis=-1)(os)
o = keras.layers.Conv2D(256, kernel_size=3, strides=1, padding="SAME")(o)
else:
raise Exception("wrong version")
return keras.Model(inputs=inputs, outputs=o)
def build(self, input_shape):
assert input_shape[0] >= 32
self.M = keras.Sequential(
[keras.layers.InputLayer(input_shape=(self.latent_dim, ))])
for _ in range(5):
self.M.add(
keras.layers.Dense(self.latent_dim,
kernel_initializer="he_normal"))
self.M.add(keras.layers.LeakyReLU(0.2))
def up(i, style, noise, l, l_):
o = keras.layers.UpSampling2D(size=l_ // l,
interpolation="bilinear")(i)
o = Conv2DMod(layer_dict[l_] * self.filter_num,
kernel_size=3,
strides=1)([o, style])
noise_crop = keras.layers.Cropping2D(
(input_shape[0] - l_) // 2)(noise)
o_n = conv(layer_dict[l_] * self.filter_num,
kernel_size=1,
strides=1)(noise_crop)
o = keras.layers.Add()([o, o_n])
o = act_layer(o)
o = Conv2DMod(layer_dict[l_] * self.filter_num,
kernel_size=3,
strides=1)([o, style])
o_n = conv(layer_dict[l_] * self.filter_num,
kernel_size=1,
strides=1)(noise_crop)
o = keras.layers.Add()([o, o_n])
o = act_layer(o)
return o
def down(i, style, noise, l, l_):
o = Conv2DMod(layer_dict[l_] * self.filter_num,
kernel_size=3,
strides=l // l_)([i, style])
noise_crop = keras.layers.Cropping2D(
(input_shape[0] - l_) // 2)(noise)
o_n = conv(layer_dict[l_] * self.filter_num,
kernel_size=1,
strides=1)(noise_crop)
o = keras.layers.Add()([o, o_n])
o = act_layer(o)
o = Conv2DMod(layer_dict[l_] * self.filter_num,
kernel_size=3,
strides=1)([o, style])
o_n = conv(layer_dict[l_] * self.filter_num,
kernel_size=1,
strides=1)(noise_crop)
o = keras.layers.Add()([o, o_n])
o = act_layer(o)
return o
def c(i, style, noise, l, l_):
o = Conv2DMod(layer_dict[l_] * self.filter_num,
kernel_size=3,
strides=1)([i, style])
noise_crop = keras.layers.Cropping2D(
(input_shape[0] - l_) // 2)(noise)
o_n = conv(layer_dict[l_] * self.filter_num,
kernel_size=1,
strides=1)(noise_crop)
o = keras.layers.Add()([o, o_n])
o = act_layer(o)
o = Conv2DMod(layer_dict[l_] * self.filter_num,
kernel_size=3,
strides=1)([o, style])
o_n = conv(layer_dict[l_] * self.filter_num,
kernel_size=1,
strides=1)(noise_crop)
o = keras.layers.Add()([o, o_n])
o = act_layer(o)
return o
style_list = []
inp = keras.layers.Input(shape=(self.latent_dim, ))
noise = keras.layers.Input(shape=input_shape)
outs = []
o_dict = {}
l_list = [min(layer_dict.keys())]
for l, s in layer_dict.items():
if l == min(layer_dict.keys()):
style = keras.layers.Input(shape=(self.latent_dim, ))
style_list.append(style)
o = keras.layers.Dense(self.latent_dim)(inp)
o = o[:, tf.newaxis, tf.newaxis, :]
o = up(o, style, noise, 1, l)
o_dict[min(layer_dict.keys())] = o
if l <= input_shape[0]:
l_list.append(l)
t_dict = dict()
for l in l_list[:-1]:
for l_ in l_list:
t_dict.setdefault(l_, [])
style = keras.layers.Input(shape=(self.latent_dim, ))
style_list.append(style)
if l == l_:
o = c(o_dict[l], style, noise, l, l_)
t_dict[l_].append(o)
elif l > l_:
o = down(o_dict[l], style, noise, l, l_)
t_dict[l_].append(o)
else:
o = up(o_dict[l], style, noise, l, l_)
t_dict[l_].append(o)
for l, os in t_dict.items():
if len(os) == 1:
o = os[0]
else:
o = keras.layers.Add()(os)
o = conv(filters=layer_dict[l] * self.filter_num,
kernel_size=1,
strides=1)(o)
o = norm_layer(o)
o = act_layer(o)
o_dict[l] = o
for l, o in o_dict.items():
style = keras.layers.Input(shape=(self.latent_dim, ))
style_list.append(style)
rgb = Conv2DMod(input_shape[-1],
kernel_size=1,
strides=1,
demod=False)([o, style])
rgb = keras.layers.UpSampling2D(size=input_shape[1] // l,
interpolation="bilinear")(rgb)
outs.append(rgb)
o = keras.layers.Add()(outs)
self._gen = keras.Model(inputs=[inp, noise] + style_list, outputs=o)
style_list_n = []
o_list = []
inp = keras.layers.Input(shape=(self.latent_dim, ))
noise = keras.layers.Input(shape=input_shape)
self.n = len(style_list)
self.m = self.n
for _ in style_list:
style = keras.layers.Input(shape=(self.latent_dim, ))
style_list_n.append(style)
o_list.append(self.M(style))
o = self._gen([inp, noise] + o_list)
self.gen = keras.Model(inputs=[inp, noise] + style_list_n, outputs=o)
def up(i, l, l_):
o = convt(filters=layer_dict[l_] * self.filter_num,
kernel_size=3,
strides=l_ // l)(i)
o = norm_layer(o)
o = act_layer(o)
return o
def down(i, l, l_):
o = conv(filters=layer_dict[l_] * self.filter_num,
kernel_size=3,
strides=l // l_)(i)
o = norm_layer(o)
o = act_layer(o)
return o
def c(i, l, l_):
o = conv(filters=layer_dict[l_] * self.filter_num,
kernel_size=3,
strides=1)(i)
o = norm_layer(o)
o = act_layer(o)
return o
images = keras.layers.Input(shape=input_shape)
o = AugmentLayer()(images)
o_dict = dict()
l_list = []
for l, s in reversed(layer_dict.items()):
if l == input_shape[0]:
o = conv(filters=layer_dict[l] * self.filter_num,
kernel_size=3,
strides=1)(o)
o = norm_layer(o)
o = act_layer(o)
o_dict[l] = o
l_list.append(l)
if l < input_shape[0]:
l_list.append(l)
t_dict = dict()
for l in l_list[:-1]:
for l_ in l_list:
t_dict.setdefault(l_, [])
if l == l_:
o = c(o_dict[l], l, l_)
# o = o_dict[l]
t_dict[l_].append(o)
elif l > l_:
o = down(o_dict[l], l, l_)
t_dict[l_].append(o)
else:
o = up(o_dict[l], l, l_)
t_dict[l_].append(o)
for l, os in t_dict.items():
if len(os) == 1:
o = os[0]
else:
o = keras.layers.Add()(os)
o = conv(filters=64, kernel_size=1, strides=1)(o)
o = norm_layer(o)
o = act_layer(o)
o = conv(filters=64, kernel_size=3, strides=1)(o)
o = norm_layer(o)
o = act_layer(o)
o = conv(filters=64, kernel_size=1, strides=1)(o)
o = norm_layer(o)
o = act_layer(o)
o_dict[l] = o
os = []
for l, o in o_dict.items():
if l >= 32:
o = conv(filters=input_shape[-1], kernel_size=1, strides=1)(o)
o = norm_layer(o)
o = act_layer(o)
os.append(o)
o = keras.layers.Flatten()(o_dict[min(layer_dict.keys())])
self.dis = keras.Model(inputs=images, outputs=[o] + os)
self.perform_gp = True
self.perform_pl = True
self.pl_mean = 0
self.pl_length = 0.
