def CapsNet(input_shape, n_class, routings):
x = layers.Input(shape=input_shape)
conv1 = layers.Conv2D(filters=64, kernel_size=(1,12), strides=(1,1), padding='same',dilation_rate = 5)(x)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.Conv2D(filters=64, kernel_size=(1,12), strides=(1,2), padding='same',dilation_rate = 1)(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.MaxPooling2D((1, 2), strides=(1, 2))(conv1)
conv1 = layers.Conv2D(filters=96, kernel_size=(1,9), strides=1, padding='same',dilation_rate = 4)(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.Conv2D(filters=96, kernel_size=(1,9), strides=1, padding='same',dilation_rate = 4)(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.MaxPooling2D((1, 2), strides=(1, 2))(conv1)
conv1 = layers.Conv2D(filters=128, kernel_size=(1,6), strides=1, padding='same',dilation_rate = 3)(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.Conv2D(filters=128, kernel_size=(1,6), strides=1, padding='same',dilation_rate = 3)(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.MaxPooling2D((1, 2), strides=(1, 2))(conv1)
conv1 = layers.Conv2D(filters=192, kernel_size=(1,3), strides=1, padding='same',dilation_rate = 2)(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.Conv2D(filters=192, kernel_size=(1,3), strides=1, padding='same',dilation_rate = 2)(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.Conv2D(filters=192, kernel_size=(1,3), strides=1, padding='same',dilation_rate = 2)(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.MaxPooling2D((1, 2), strides=(1, 2))(conv1)
conv1 = layers.Conv2D(filters=256, kernel_size=(1,3), strides=1, padding='same',dilation_rate = 2)(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.Conv2D(filters=256, kernel_size=(1,3), strides=1, padding='same',dilation_rate = 2)(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.Conv2D(filters=256, kernel_size=(1,3), strides=1, padding='same',dilation_rate = 2)(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
primarycaps = PrimaryCap(conv1, dim_capsule=8, n_channels=32, kernel_size=(1,3),
strides=1, padding='same')
digitcaps = CapsuleLayer(num_capsule=n_class, dim_capsule=args.dim_capsule, routings=routings,
name='digitcaps')(primarycaps)
out_caps = Length(name='capsnet')(digitcaps)
model = models.Model(x, out_caps)
return model
def CapsNet(input_shape, n_class, routings):
x = layers.Input(shape=input_shape)
conv1 = layers.Conv2D(filters=256,
kernel_size=9,
strides=1,
padding='valid',
activation='relu',
name='conv1')(x)
primarycaps = PrimaryCap(conv1,
dim_capsule=8,
n_channels=32,
kernel_size=9,
strides=2,
padding='valid')
digitcaps = CapsuleLayer(num_capsule=n_class,
dim_capsule=16,
routings=routings,
name='digitcaps')(primarycaps)
out_caps = Length(name='capsnet')(digitcaps)
y = layers.Input(shape=(2, ))
masked_by_y = Mask()([digitcaps, y])
masked_by_y0 = Lambda(lambda x: x[:, 0, :])(masked_by_y)
masked_by_y1 = Lambda(lambda x: x[:, 1, :])(masked_by_y)
masked = Mask()(digitcaps)
masked0 = Lambda(lambda x: x[:, 0, :])(masked)
masked1 = Lambda(lambda x: x[:, 1, :])(masked)
decoder = models.Sequential(name='decoder')
decoder.add(layers.Reshape(target_shape=(4, 4, 1), input_shape=(16, )))
decoder.add(layers.ZeroPadding2D(((3, 0), (3, 0))))
decoder.add(layers.UpSampling2D((2, 2)))
decoder.add(layers.Conv2D(256, 3, activation='relu', padding='same'))
decoder.add(layers.Conv2D(256, 3, activation='relu', padding='same'))
decoder.add(layers.UpSampling2D((2, 2)))
decoder.add(layers.Conv2D(128, 3, activation='relu', padding='same'))
decoder.add(layers.Conv2D(128, 3, activation='relu', padding='same'))
decoder.add(layers.Conv2D(1, 1))
train_model = models.Model(
[x, y],
[out_caps, decoder(masked_by_y0),
decoder(masked_by_y1)])
eval_model = models.Model(
x, [out_caps, decoder(masked0),
decoder(masked1)])
return train_model, eval_model
def CapsNet(input_shape, n_class, routings):
x = layers.Input(shape=input_shape)
conv1 = layers.Conv2D(filters=64, kernel_size=(1,3), strides=(1,1), padding='same')(x)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.Conv2D(filters=64, kernel_size=(1,3), strides=(1,1), padding='same')(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.MaxPooling2D((1, 2), strides=(1, 2))(conv1)
conv1 = layers.Conv2D(filters=96, kernel_size=(1,3), strides=1, padding='same')(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.Conv2D(filters=96, kernel_size=(1,3), strides=1, padding='same')(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.MaxPooling2D((1, 2), strides=(1, 2))(conv1)
conv1 = layers.Conv2D(filters=128, kernel_size=(1,3), strides=1, padding='same')(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.Conv2D(filters=128, kernel_size=(1,3), strides=1, padding='same')(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.MaxPooling2D((1, 2), strides=(1, 2))(conv1)
conv1 = layers.Conv2D(filters=192, kernel_size=(1,3), strides=1, padding='same')(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.Conv2D(filters=192, kernel_size=(1,3), strides=1, padding='same')(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.MaxPooling2D((1, 2), strides=(1, 2))(conv1)
conv1 = layers.Conv2D(filters=256, kernel_size=(1,3), strides=1, padding='same')(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
conv1 = layers.Conv2D(filters=256, kernel_size=(1,3), strides=1, padding='same')(conv1)
conv1 = ELU(alpha=0.5)(conv1)
conv1 = BN()(conv1)
primarycaps = PrimaryCap(conv1, dim_capsule=8, n_channels=32, kernel_size=(1,3),
strides=1, padding='same')
digitcaps = CapsuleLayer(num_capsule=n_class, dim_capsule=args.dim_capsule, routings=routings,
name='digitcaps')(primarycaps)
out_caps = Length(name='capsnet')(digitcaps)
y = layers.Input(shape=(2,))
masked_by_y = Mask()([digitcaps, y])
masked_by_y0 = Lambda(lambda x: x[:,0,:])(masked_by_y)
masked_by_y1 = Lambda(lambda x: x[:,1,:])(masked_by_y)
masked = Mask()(digitcaps)
masked0 = Lambda(lambda x: x[:,0,:])(masked)
masked1 = Lambda(lambda x: x[:,1,:])(masked)
decoder = models.Sequential(name='decoder')
decoder.add(layers.Dense(375, input_dim=args.dim_capsule))
decoder.add(ELU(alpha=0.5))
decoder.add(layers.Reshape(target_shape=(1,375,1), name='out_recon'))
decoder.add(UpSampling2D((1,2)))
decoder.add(layers.Conv2D(filters=256, kernel_size=(1,3), strides=1, padding='same'))
decoder.add(ELU(alpha=0.5))
decoder.add(BN())
decoder.add(layers.Conv2D(filters=256, kernel_size=(1,3), strides=1, padding='same'))
decoder.add(ELU(alpha=0.5))
decoder.add(BN())
decoder.add(UpSampling2D((1,2)))
decoder.add(layers.Conv2D(filters=128, kernel_size=(1,3), strides=1, padding='same'))
decoder.add(ELU(alpha=0.5))
decoder.add(BN())
decoder.add(layers.Conv2D(filters=128, kernel_size=(1,3), strides=1, padding='same'))
decoder.add(ELU(alpha=0.5))
decoder.add(BN())
decoder.add(UpSampling2D((1,2)))
decoder.add(layers.Conv2D(filters=64, kernel_size=(1,3), strides=1, padding='same'))
decoder.add(ELU(alpha=0.5))
decoder.add(BN())
decoder.add(layers.Conv2D(filters=64, kernel_size=(1,3), strides=1, padding='same'))
decoder.add(ELU(alpha=0.5))
decoder.add(BN())
decoder.add(layers.Conv2D(filters=1, kernel_size=(1,1), strides=1, padding='same'))
train_model = models.Model([x, y], [out_caps, decoder(masked_by_y0), decoder(masked_by_y1)])
eval_model = models.Model(x, [out_caps, decoder(masked0), decoder(masked1)])
return train_model, eval_model
def CapsNet(input_shape, n_class, num_routing):
from keras import layers, models
from capsulelayers2 import CapsuleLayer, PrimaryCap, Length, Mask
from keras.preprocessing import sequence
"""
A Capsule Network on MNIST.
:param input_shape: data shape, 4d, [None, width, height, channels]
:param n_class: number of classes
:param num_routing: number of routing iterations
:return: A Keras Model with 2 inputs and 2 outputs
"""
x = layers.Input(shape=input_shape)
# conv1 = layers.Conv1D(filters=256, kernel_size=9, strides=1, padding='valid', activation='relu', name='conv1')(embed)
conv1 = layers.Conv2D(filters=args.filters,
kernel_size=(4, args.kernel_size),
strides=(1, 1),
padding='valid',
activation='relu',
name='conv1')(x)
# conv1 = layers.Dropout(0.1)(conv1)
# Layer 2: Conv2D layer with `squash` activation, then reshape to [None, num_capsule, dim_vector]
# primarycaps = PrimaryCap(conv1, dim_vector=8, n_channels=32, kernel_size=9, strides=2, padding='valid')
primarycaps = PrimaryCap(conv1,
dim_vector=8,
n_channels=32,
kernel_size=(1, 9),
strides=2,
padding='valid')
# Layer 3: Capsule layer. Routing algorithm works here.
# digitcaps = CapsuleLayer(num_capsule=n_class, dim_vector=16, num_routing=num_routing, name='digitcaps')(primarycaps)
digitcaps = CapsuleLayer(num_capsule=n_class,
dim_vector=16,
num_routing=args.rout,
name='digitcaps')(primarycaps)
# Layer 4: This is an auxiliary layer to replace each capsule with its length. Just to match the true label's shape.
# If using tensorflow, this will not be necessary. :)
out_caps = Length(name='out_caps')(digitcaps)
# Decoder network.
y = layers.Input(shape=(n_class, ))
masked_by_y = Mask()(
[digitcaps, y]
) # The true label is used to mask the output of capsule layer. For training
masked = Mask(
)(digitcaps) # Mask using the capsule with maximal length. For prediction
# Shared Decoder model in training and prediction
decoder = models.Sequential(name='decoder')
decoder.add(layers.Dense(512, activation='relu', input_dim=16 * n_class))
decoder.add(layers.Dense(1024, activation='relu'))
decoder.add(layers.Dense(np.prod(input_shape), activation='sigmoid'))
decoder.add(layers.Reshape(target_shape=input_shape, name='out_recon'))
# Models for training and evaluation (prediction)
train_model = models.Model([x, y], [out_caps, decoder(masked_by_y)])
eval_model = models.Model(x, [out_caps, decoder(masked)])
# manipulate model
noise = layers.Input(shape=(n_class, 16))
noised_digitcaps = layers.Add()([digitcaps, noise])
masked_noised_y = Mask()([noised_digitcaps, y])
manipulate_model = models.Model([x, y, noise], decoder(masked_noised_y))
return train_model, eval_model, manipulate_model