def build_segnet(self):
img_w = self.width
img_h = self.height
n_labels = 2
kernel = (3, 3)
encoding_layers = [
Conv2D(64, kernel, padding='same', input_shape=(img_h, img_w, 3)),
BatchNormalization(),
Activation('relu'),
Conv2D(64, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
MaxPool2D(),
Conv2D(128, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(128, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
MaxPool2D(),
Conv2D(256, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(256, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(256, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
MaxPool2D(),
Conv2D(512, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(512, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(512, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
MaxPool2D(),
Conv2D(512, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(512, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(512, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
MaxPool2D(),
]
autoencoder = Sequential()
autoencoder.encoding_layers = encoding_layers
for l in autoencoder.encoding_layers:
autoencoder.add(l)
print(l.input_shape, l.output_shape, l)
decoding_layers = [
UpSampling2D(),
Conv2D(512, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(512, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(512, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
UpSampling2D(),
Conv2D(512, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(512, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(256, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
UpSampling2D(),
Conv2D(256, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(256, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(128, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
UpSampling2D(),
Conv2D(128, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(64, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
UpSampling2D(),
Conv2D(64, kernel, padding='same'),
BatchNormalization(),
Activation('relu'),
Conv2D(n_labels, (1, 1), padding='valid'),
BatchNormalization(),
]
autoencoder.decoding_layers = decoding_layers
for l in autoencoder.decoding_layers:
autoencoder.add(l)
print(l.input_shape, l.output_shape, l)
autoencoder.add(Reshape((n_labels, img_h * img_w)))
autoencoder.add(Permute((2, 1)))
autoencoder.add(Reshape((img_h, img_w, n_labels)))
autoencoder.add(Activation('softmax'))
return autoencoder
