def build_model(encoder_dim=50, bottleneck_dim=20):
encoder = Dense(encoder_dim, input_dim=28 * 28, activation='tanh')
bottleneck = Dense(bottleneck_dim, activation='tanh')
bottleneck_2 = DependentDense(encoder_dim, bottleneck, activation='tanh')
decoder = DependentDense(28 * 28, encoder, activation='tanh')
model = models.Sequential()
model.add(encoder)
model.add(bottleneck)
model.add(bottleneck_2)
model.add(decoder)
return model, encoder
def build_model(nb_filters=32, nb_pool=2, nb_conv=3):
model = models.Sequential()
d = Dense(30)
c = Convolution2D(nb_filters,
nb_conv,
nb_conv,
border_mode='same',
input_shape=(1, 28, 28))
# ========= ENCODER ========================
model.add(c)
model.add(Activation('tanh'))
model.add(MaxPooling2D(pool_size=(nb_pool, nb_pool)))
model.add(Dropout(0.25))
# ========= BOTTLENECK ======================
model.add(Flatten())
model.add(d)
model.add(Activation('tanh'))
# ========= BOTTLENECK^-1 =====================
model.add(DependentDense(nb_filters * 14 * 14, d))
model.add(Activation('tanh'))
model.add(Reshape((nb_filters, 14, 14)))
# ========= DECODER =========================
model.add(UpSampling2D(size=(nb_pool, nb_pool)))
model.add(Deconvolution2D(c, border_mode='same'))
model.add(Activation('tanh'))
return model
def build_model(nb_filters=32, nb_pool=2, nb_conv=2):
model = models.Sequential()
d = Dense(60)
c = Convolution2D(nb_filters,
nb_conv,
nb_conv,
border_mode='same',
input_shape=(28, 28, 1))
mp = MaxPooling2D(pool_size=(nb_pool, nb_pool))
# ========= ENCODER ========================
model.add(c)
model.add(Activation('relu'))
model.add(mp)
model.add(Dropout(0.25))
# ========= BOTTLENECK ======================
model.add(Flatten())
model.add(d)
model.add(Activation('relu'))
# ========= BOTTLENECK^-1 =====================
model.add(DependentDense(units=nb_filters * 14 * 14, master_layer=d))
model.add(Activation('relu'))
model.add(Reshape((nb_filters, 14, 14)))
# ========= DECODER =========================
model.add(DePool2D(mp, size=(nb_pool, nb_pool)))
model.add(
Deconvolution2D(binded_conv_layer=c,
border_mode='same',
filters=nb_filters,
kernel_size=(nb_conv, nb_conv)))
model.add(Activation('relu'))
return model
def get_decoder(C_1, C_2, C_3, c, c2, d, mp, mp2, nb_pool):
decoder = models.Sequential()
decoder.add(
DependentDense(d.input_shape[1], d, input_shape=(d.output_shape[1], )))
decoder.add(Activation('tanh'))
decoder.add(Reshape((C_2, 8, 8)))
# ====================================================
# decoder.add(DePool2D(mp3, size=(nb_pool, nb_pool)))
# decoder.add(Deconvolution2D(c3, nb_out_channels=C_2, border_mode='same'))
# decoder.add(Activation('tanh'))
# ====================================================
decoder.add(DePool2D(mp2, size=(nb_pool, nb_pool)))
decoder.add(Deconvolution2D(c2, nb_out_channels=C_1, border_mode='same'))
decoder.add(Activation('tanh'))
# ====================================================
decoder.add(DePool2D(mp, size=(nb_pool, nb_pool)))
decoder.add(Deconvolution2D(c, nb_out_channels=3, border_mode='same'))
decoder.add(Activation('tanh'))
# ====================================================
return decoder
def build_model(nb_filters=32, nb_pool=2, nb_conv=3):
model = models.Sequential()
d = Dense(30)
c = Convolution2D(nb_filters, (nb_conv, nb_conv), padding='same', input_shape=(28, 28, 1))
mp =MaxPooling2D(pool_size=(nb_pool, nb_pool), padding='same')
# ========= ENCODER ========================
model.add(c)
model.add(Activation('tanh'))
model.add(mp)
model.add(Dropout(0.25))
# ========= BOTTLENECK ======================
model.add(Flatten())
model.add(d)
model.add(Activation('tanh'))
# ========= BOTTLENECK^-1 =====================
model.add(DependentDense(nb_filters * 14 * 14, d))
model.add(Activation('tanh'))
model.add(Reshape((14, 14, nb_filters)))
# ========= DECODER =========================
model.add(DePool2D(mp, size=(nb_pool, nb_pool)))
model.add(Conv2DTranspose(1, (nb_conv, nb_conv), padding='same'))
model.add(Activation('tanh'))
return model
def build_model(nb_filters=32, nb_pool=2, nb_conv=3):
C_1 = 64
C_2 = 32
C_3 = 16
c1 = Convolution2D(C_1,
nb_conv,
nb_conv,
border_mode='same',
name='c1',
input_shape=(3, 32, 32))
mp1 = MaxPooling2D(pool_size=(nb_pool, nb_pool), name='mp1')
c2 = Convolution2D(C_2, nb_conv, nb_conv, border_mode='same', name='c2')
mp2 = MaxPooling2D(pool_size=(nb_pool, nb_pool), name='mp2')
c3 = Convolution2D(C_3, nb_conv, nb_conv, border_mode='same', name='c3')
mp3 = MaxPooling2D(pool_size=(nb_pool, nb_pool), name='mp3')
d = Dense(100, name='encoded')
model = Sequential()
# ====================================================
model.add(c1)
model.add(Activation('tanh'))
model.add(mp1)
# ====================================================
model.add(Dropout(0.25))
# ====================================================
model.add(c2)
model.add(Activation('tanh'))
model.add(mp2)
# ====================================================
model.add(c3)
model.add(Activation('tanh'))
model.add(mp3)
# ====================================================
model.add(Dropout(0.25))
# ====================================================
model.add(Flatten())
model.add(d)
model.add(Activation('tanh'))
# ====================================================
model.add(
DependentDense(d.input_shape[1], d, input_shape=(d.output_shape[1], )))
model.add(Activation('tanh'))
model.add(Reshape((C_3, 4, 4)))
# ====================================================
model.add(DePool2D(mp3, size=(nb_pool, nb_pool)))
model.add(Deconvolution2D(c3, nb_out_channels=C_2, border_mode='same'))
model.add(Activation('tanh'))
# ====================================================
model.add(DePool2D(mp2, size=(nb_pool, nb_pool)))
model.add(Deconvolution2D(c2, nb_out_channels=C_1, border_mode='same'))
model.add(Activation('tanh'))
# ====================================================
model.add(DePool2D(mp1, size=(nb_pool, nb_pool)))
model.add(Deconvolution2D(c1, nb_out_channels=3, border_mode='same'))
model.add(Activation('tanh'))
# ====================================================
model.compile('adam', loss='mean_squared_error')
#model.compile('rmsprop', loss='mean_squared_error')
return model