base = VGG16(weights='imagenet', include_top=False, input_shape=(64,64,3))

# convert to sequential model
model = Sequential()
for layer in base.layers:
    model.add(layer)

# Remove last layer
model.layers.pop()

# add flatten and two dense layers that don't appear when specifying an input_shape
model.add(Flatten())
model.add(Dense(4096))
model.add(Dropout(.5))
model.add(Dense(4096))
model.add(Dropout(.5))

for layer in model.layers:
    layer.trainable = False

# Add a layer for 3 classes
model.add(Dense(3, activation='softmax'))

model.compile(
    optimizer = 'rmsprop',
    loss='categorical_crossentropy',
    metrics=['accuracy']
)

# training
model.fit(x_train,y_train, epochs=30, batch_size=64, verbose=1)

# predict
y_target = model.predict(x_target, batch_size=64, verbose=1)