ASAC 51일차_딥러닝 6일차

2024.02.14

VGG 모델

CNN 구조를 가지는 VGG 모델 적용

실습

VGG16 관련 모델 논문을 참고하여 Fashion-MNIST 셋에 맞추어 튜닝

model_vgg_fshion= tf.keras.Sequential([
    # 원본 입력 : 224 224 RGB  ===> 28, 28, 1
    Conv2D(input_shape=(28,28,1),kernel_size=(3,3),
           filters=64, padding="same", activation="relu" ),
    Conv2D(kernel_size=(3,3),filters=64, padding="same", activation="relu"),
    MaxPool2D(pool_size = (2,2), strides=(2,2)),

    Conv2D(kernel_size=(3,3), filters=128, padding="same", activation="relu"),
    Conv2D(kernel_size=(3,3), filters=128, padding="same", activation="relu"),
    MaxPool2D(pool_size = (2,2), strides=(2,2)),

    Conv2D(kernel_size=(3,3), filters=256, padding="same", activation="relu"),
    Conv2D(kernel_size=(3,3), filters=256, padding="same", activation="relu"),
    Conv2D(kernel_size=(3,3), filters=256, padding="same", activation="relu"),
    MaxPool2D(pool_size = (2,2), strides=(2,2)),

    # 분류를 위한 FNN
    # 1) 특징에 대한 Flatten작업
    Flatten(),
    
    # 2) 분류용 HL
    Dense( units=4096, activation="relu"),
    Dense( units=4096, activation="relu"),

    # 3) 출력용 ==> 10종류
    Dense(units=10, activation="softmax"),
	])
    
model_vgg_fshion.summary()
#----------------------------------------------------------#
# callback 적용

import os
cp_path = "training/cp-{epoch:04d}.ckpt"
cp_dir  = os.path.dirname(cp_path)
cp_callback = tf.keras.callbacks.ModelCheckpoint(
    cp_path,
    verbose = 1,
    save_weights_only = True
    )
    
es_callback = tf.keras.callbacks.EarlyStopping(
    monitor = "val_loss", 
    patience = 10
)
#----------------------------------------------------------#
model_vgg_fshion.compile(
    optimizer = tf.keras.optimizers.Adam(),
    loss = "sparse_categorical_crossentropy",
    metrics = ["accuracy"]
)
model_vgg_fshion.fit(train_X, train_y,
                     epochs=100,
                     validation_split=0.25,
                     batch_size = 1024,
                     callbacks=[cp_callback,es_callback ]
                     )