Untitled

#  dropout rate for FC layers
dropout=0.5

# CNN architecture

input_image = Input(shape=(img_height,img_width,img_channels))
x1 = Conv2D(64, (3, 3),padding='same', activation='relu')(input_image)
x1 = Conv2D(64, (3, 3),padding='same', activation='relu')(x1)
x1 = MaxPooling2D((2, 2))(x1)

x2_pr = Conv2D(128, (3, 3),padding='same', activation='relu')(x1)
x2 = Conv2D(128, (3, 3),padding='same', activation='relu')(x2_pr)


x2 = MaxPooling2D((2, 2))(x2)
x3_pr = Conv2D(256, (3, 3),padding='same', activation='relu')(x2)
x3 = Conv2D(256, (3, 3),padding='same', activation='relu')(x3_pr)
x3 = Conv2D(256, (1, 1),padding='same',activation='relu')(x3)



x3 = MaxPooling2D((2, 2))(x3)
x4_pr = Conv2D(512, (3, 3),padding='same', activation='relu')(x3)
x4 = Conv2D(512, (3, 3),padding='same', activation='relu')(x4_pr)
x4 = Conv2D(512, (1, 1),padding='same',activation='relu')(x4)


x4 = MaxPooling2D((2, 2))(x4)
x5_pr = Conv2D(512, (3, 3),padding='same', activation='relu')(x4)
x5 = Conv2D(512, (3, 3),padding='same', activation='relu')(x5_pr)
x5 = Conv2D(512, (1, 1),padding='same',activation='relu')(x5)



x5 = MaxPooling2D((2, 2))(x5)

x5 = Flatten()(x5)

x=Dense(4096, activation='relu', kernel_constraint=maxnorm(3))(x5)
x=Dropout(dropout)(x)
x=Dense(4096, activation='relu', kernel_constraint=maxnorm(3))(x)
x=Dropout(dropout)(x)

out= Dense(num_classes, activation='softmax')(x)

model = Model(inputs = input_image, outputs = out);