[딥러닝] VGG16 모델

from tensorflow.keras.applications import VGG16

# https://www.tensorflow.org/api_docs/python/tf/keras/applications/vgg16/VGG16?hl=ko
# 여기 위에서 해당 모델 부분 확인 가능함 

pre_trainde_model =VGG16(input_shape=(224,224,3),
                        
                         include_top=False,
                         weights ='imagenet'
                         
                         )

 # include_top =False : 불러온 모델의 MLP층을 사용하지 않고 특성추출부만 사용 (특성추출방식)
 # > 기존 이미지넷 대회에서는 1000가지의 이미지를 분류했으나 우리는 3가지 동물로만 분류하기 때문에
 # MLP층을 우리에게 맞게 설정해줘야함
 # weights ='imagenet' : 이미지넷 대회에서 학습된 가중치(w)를 그대로 가져와서 사용 
 
cnn_model2 = Sequential()

# 특성추출부
cnn_model2.add(pre_trainde_model)

# MLP층
cnn_model2.add(Flatten())
cnn_model2.add(Dense(128, activation='relu'))
cnn_model2.add(Dense(64, activation='relu'))
cnn_model2.add(Dense(32, activation='relu'))

cnn_model2.add(Dense(3, activation='softmax'))

cnn_model2.summary()

Model: "sequential_1"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 vgg16 (Functional)          (None, 7, 7, 512)         14714688  
                                                                 
 flatten_1 (Flatten)         (None, 25088)             0         
                                                                 
 dense_4 (Dense)             (None, 128)               3211392   
                                                                 
 dense_5 (Dense)             (None, 64)                8256      
                                                                 
 dense_6 (Dense)             (None, 32)                2080      
                                                                 
 dense_7 (Dense)             (None, 3)                 99        
                                                                 
=================================================================
Total params: 17,936,515
Trainable params: 17,936,515
Non-trainable params: 0
_________________________________________________________________

 

# 사전에 학습된 vgg16 모델을 불러와서 재학습이 불가능하도록 동결(잘 학습된 w,b 값을 그대로 사용하기 위해서)
# > 동결을 시지키 않게 되면 새롭게 만들어준 MLP층과 재학습하게 되므로 잘 학습된 w,b값이 뒤틀려 버라게 됨

pre_trainde_model.trainable = False

pre_trainde_model.summary()

Model: "vgg16"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_1 (InputLayer)        [(None, 224, 224, 3)]     0         
                                                                 
 block1_conv1 (Conv2D)       (None, 224, 224, 64)      1792      
                                                                 
 block1_conv2 (Conv2D)       (None, 224, 224, 64)      36928     
                                                                 
 block1_pool (MaxPooling2D)  (None, 112, 112, 64)      0         
                                                                 
 block2_conv1 (Conv2D)       (None, 112, 112, 128)     73856     
                                                                 
 block2_conv2 (Conv2D)       (None, 112, 112, 128)     147584    
                                                                 
 block2_pool (MaxPooling2D)  (None, 56, 56, 128)       0         
                                                                 
 block3_conv1 (Conv2D)       (None, 56, 56, 256)       295168    
                                                                 
 block3_conv2 (Conv2D)       (None, 56, 56, 256)       590080    
                                                                 
 block3_conv3 (Conv2D)       (None, 56, 56, 256)       590080    
                                                                 
 block3_pool (MaxPooling2D)  (None, 28, 28, 256)       0         
                                                                 
 block4_conv1 (Conv2D)       (None, 28, 28, 512)       1180160   
                                                                 
 block4_conv2 (Conv2D)       (None, 28, 28, 512)       2359808   
                                                                 
 block4_conv3 (Conv2D)       (None, 28, 28, 512)       2359808   
                                                                 
 block4_pool (MaxPooling2D)  (None, 14, 14, 512)       0         
                                                                 
 block5_conv1 (Conv2D)       (None, 14, 14, 512)       2359808   
                                                                 
 block5_conv2 (Conv2D)       (None, 14, 14, 512)       2359808   
                                                                 
 block5_conv3 (Conv2D)       (None, 14, 14, 512)       2359808   
                                                                 
 block5_pool (MaxPooling2D)  (None, 7, 7, 512)         0         
                                                                 
=================================================================
Total params: 14,714,688
Trainable params: 0
Non-trainable params: 14,714,688
_________________________________________________________________

 

 

cnn_model2.compile(loss='sparse_categorical_crossentropy',
                   optimizer='Adam',
                   metrics =['acc']
                   )
                   
from sklearn.utils import validation
h2= cnn_model2.fit(X_train, y_train, 
                   validation_split=0.2,
                   batch_size=128,
                   epochs=50
                   )
                   
plt.figure(figsize=(15,10))

plt.plot(h2.history['acc'],
         label='acc',
         c = 'blue',
         marker='.'
         )

plt.plot(h2.history['val_acc'],
         label='val_sigmid+SGD',
         c = 'red',
         marker='.'
         )

plt.legend()
plt.show()

 

pre = cnn_model2.predict(X_test)

print(classification_report(y_test, np.argmax(pre, axis=1)))

              precision    recall  f1-score   support

           0       0.77      0.77      0.77        87
           1       0.85      0.66      0.74        79
           2       0.66      0.82      0.73        74

    accuracy                           0.75       240
   macro avg       0.76      0.75      0.75       240
weighted avg       0.76      0.75      0.75       240