seti14.py

INPUT_SHAPE = (192, 192, 3)

# load the pre-trained model
vgg = keras.applications.vgg19.VGG19(include_top=False, weights='imagenet', 
                                     input_shape=INPUT_SHAPE)

vgg.trainable = True
set_trainable = False
for layer in vgg.layers:
    if layer.name in ['block5_conv1', 'block4_conv1']:
        set_trainable = True
    if set_trainable:
        layer.trainable = True
    else:
        layer.trainable = False

# add custom dense and output layers
base_vgg = vgg
base_out = base_vgg.output
pool_out = keras.layers.Flatten()(base_out)
hidden1 = keras.layers.Dense(1024, activation='relu')(pool_out)
drop1 = keras.layers.Dropout(rate=0.2)(hidden1)
hidden2 = keras.layers.Dense(512, activation='relu')(drop1)
drop2 = keras.layers.Dropout(rate=0.2)(hidden2)

out = keras.layers.Dense(7, activation='softmax')(drop2)

model = keras.Model(inputs=base_vgg.input, outputs=out)
model.compile(optimizer=keras.optimizers.RMSprop(lr=1e-5),
                loss='categorical_crossentropy',
                metrics=[categorical_accuracy])
model.summary()

# Output
Layer (type)                 Output Shape              Param #   
=================================================================
input_2 (InputLayer)         (None, 192, 192, 3)       0         
_________________________________________________________________
block1_conv1 (Conv2D)        (None, 192, 192, 64)      1792      
_________________________________________________________________
block1_conv2 (Conv2D)        (None, 192, 192, 64)      36928     
_________________________________________________________________
block1_pool (MaxPooling2D)   (None, 96, 96, 64)        0         
_________________________________________________________________
...
...
_________________________________________________________________
dense_5 (Dense)              (None, 512)               524800    
_________________________________________________________________
dropout_4 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_6 (Dense)              (None, 7)                 3591      
=================================================================
Total params: 39,428,167
Trainable params: 37,102,599
Non-trainable params: 2,325,568
_________________________________________________________________