[딥러닝] VGG16 바탕으로 신경망 구성하기, 배치 정규화, 드롭아웃 적용하기

즐거운프로그래밍

[딥러닝] VGG16 바탕으로 신경망 구성하기, 배치 정규화, 드롭아웃 적용하기

수수께끼 고양이 2023. 11. 2. 18:01

728x90

참고 링크

[딥러닝 | CNN] - VGG16

VGG16이 수록된 논문 "Very deep convolutional networks for large-scale image recognition" 의 Model Architecture 설명부분까지의 내용을 기반으로 정리하겠다. 1. VGG16란? ILSVRC 2014년 대회에서 2위를 한 CNN모델이다. 그중

brave-greenfrog.tistory.com

연습문제
1. VGG16 신경망 텐서플로로 구성하기
2. 임의 데이터 만드어서 신경망 테스트하기

import tensorflow as tf

model=tf.keras.Sequential([
    tf.keras.layers.InputLayer(input_shape=(224,224,3)),
    tf.keras.layers.Conv2D(64,(3,3),(1,1),padding='same'),
    tf.keras.layers.Conv2D(64,(3,3),(1,1),padding='same'),
    tf.keras.layers.MaxPooling2D((2,2),(2,2)),
    tf.keras.layers.Conv2D(128,(3,3),(1,1),padding='same'),
    tf.keras.layers.Conv2D(128,(3,3),(1,1),padding='same'),
    tf.keras.layers.MaxPooling2D((2,2),(2,2)),
    tf.keras.layers.Conv2D(256,(3,3),(1,1),padding='same'),
    tf.keras.layers.Conv2D(256,(3,3),(1,1),padding='same'),
    tf.keras.layers.Conv2D(256,(3,3),(1,1),padding='same'),
    tf.keras.layers.MaxPooling2D((2,2),(2,2)),
    tf.keras.layers.Conv2D(512,(3,3),(1,1),padding='same'),
    tf.keras.layers.Conv2D(512,(3,3),(1,1),padding='same'),
    tf.keras.layers.Conv2D(512,(3,3),(1,1),padding='same'),
    tf.keras.layers.MaxPooling2D((2,2),(2,2)),
    tf.keras.layers.Conv2D(512,(3,3),(1,1),padding='same'),
    tf.keras.layers.Conv2D(512,(3,3),(1,1),padding='same'),
    tf.keras.layers.Conv2D(512,(3,3),(1,1),padding='same'),
    tf.keras.layers.MaxPooling2D((2,2),(2,2)),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(4096),
    tf.keras.layers.Dense(4096),
    tf.keras.layers.Dense(1000,activation='relu')
])

model.summary()

연습문제

3. cifar10 데이터로 신경망 학습해보기
4. BatchNormalization 적용해보기

5. Dropout 적용해보기

import tensorflow as tf

mnist=tf.keras.datasets.cifar10
(X,YT),(x,yt)=mnist.load_data()
X,x=X/255,x/255

model=tf.keras.Sequential([
    tf.keras.layers.InputLayer(input_shape=(32,32,3)),
    tf.keras.layers.Conv2D(64,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Conv2D(64,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.MaxPooling2D((2,2),(2,2)),
    tf.keras.layers.Conv2D(128,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Conv2D(128,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.MaxPooling2D((2,2),(2,2)),
    tf.keras.layers.Conv2D(256,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Conv2D(256,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Conv2D(256,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.MaxPooling2D((2,2),(2,2)),
    tf.keras.layers.Conv2D(512,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Conv2D(512,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Conv2D(512,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.MaxPooling2D((2,2),(2,2)),
    tf.keras.layers.Conv2D(512,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Conv2D(512,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Conv2D(512,(3,3),(1,1),padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.MaxPooling2D((2,2),(2,2)),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(4096),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Dense(4096),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.ReLU(),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Dense(10,activation='relu')
])

model.summary()

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

model.fit(X,YT,epochs=5)

model.evaluate(x,yt)

Model: "sequential"
_________________________________________________________________
Layer (type)                Output Shape              Param #
=================================================================
conv2d (Conv2D)             (None, 32, 32, 64)        1792

batch_normalization (BatchN  (None, 32, 32, 64)       256
ormalization)

re_lu (ReLU)                (None, 32, 32, 64)        0

dropout (Dropout)           (None, 32, 32, 64)        0

conv2d_1 (Conv2D)           (None, 32, 32, 64)        36928

batch_normalization_1 (Batc  (None, 32, 32, 64)       256
hNormalization)

re_lu_1 (ReLU)              (None, 32, 32, 64)        0

dropout_1 (Dropout)         (None, 32, 32, 64)        0

max_pooling2d (MaxPooling2D  (None, 16, 16, 64)       0
)

conv2d_2 (Conv2D)           (None, 16, 16, 128)       73856

batch_normalization_2 (Batc  (None, 16, 16, 128)      512
hNormalization)

re_lu_2 (ReLU)              (None, 16, 16, 128)       0

dropout_2 (Dropout)         (None, 16, 16, 128)       0

conv2d_3 (Conv2D)           (None, 16, 16, 128)       147584

batch_normalization_3 (Batc  (None, 16, 16, 128)      512
hNormalization)

re_lu_3 (ReLU)              (None, 16, 16, 128)       0

dropout_3 (Dropout)         (None, 16, 16, 128)       0

max_pooling2d_1 (MaxPooling  (None, 8, 8, 128)        0
2D)

conv2d_4 (Conv2D)           (None, 8, 8, 256)         295168

batch_normalization_4 (Batc  (None, 8, 8, 256)        1024
hNormalization)

re_lu_4 (ReLU)              (None, 8, 8, 256)         0

dropout_4 (Dropout)         (None, 8, 8, 256)         0

conv2d_5 (Conv2D)           (None, 8, 8, 256)         590080

batch_normalization_5 (Batc  (None, 8, 8, 256)        1024
hNormalization)

re_lu_5 (ReLU)              (None, 8, 8, 256)         0

dropout_5 (Dropout)         (None, 8, 8, 256)         0

conv2d_6 (Conv2D)           (None, 8, 8, 256)         590080

batch_normalization_6 (Batc  (None, 8, 8, 256)        1024
hNormalization)

re_lu_6 (ReLU)              (None, 8, 8, 256)         0

dropout_6 (Dropout)         (None, 8, 8, 256)         0

max_pooling2d_2 (MaxPooling  (None, 4, 4, 256)        0
2D)

conv2d_7 (Conv2D)           (None, 4, 4, 512)         1180160

batch_normalization_7 (Batc  (None, 4, 4, 512)        2048
hNormalization)

re_lu_7 (ReLU)              (None, 4, 4, 512)         0

dropout_7 (Dropout)         (None, 4, 4, 512)         0

conv2d_8 (Conv2D)           (None, 4, 4, 512)         2359808

batch_normalization_8 (Batc  (None, 4, 4, 512)        2048
hNormalization)

re_lu_8 (ReLU)              (None, 4, 4, 512)         0

dropout_8 (Dropout)         (None, 4, 4, 512)         0

conv2d_9 (Conv2D)           (None, 4, 4, 512)         2359808

batch_normalization_9 (Batc  (None, 4, 4, 512)        2048
hNormalization)

re_lu_9 (ReLU)              (None, 4, 4, 512)         0

dropout_9 (Dropout)         (None, 4, 4, 512)         0

max_pooling2d_3 (MaxPooling  (None, 2, 2, 512)        0
2D)

conv2d_10 (Conv2D)          (None, 2, 2, 512)         2359808

batch_normalization_10 (Bat  (None, 2, 2, 512)        2048
chNormalization)

re_lu_10 (ReLU)             (None, 2, 2, 512)         0

dropout_10 (Dropout)        (None, 2, 2, 512)         0

conv2d_11 (Conv2D)          (None, 2, 2, 512)         2359808

batch_normalization_11 (Bat  (None, 2, 2, 512)        2048
chNormalization)

re_lu_11 (ReLU)             (None, 2, 2, 512)         0

dropout_11 (Dropout)        (None, 2, 2, 512)         0

conv2d_12 (Conv2D)          (None, 2, 2, 512)         2359808

batch_normalization_12 (Bat  (None, 2, 2, 512)        2048
chNormalization)

re_lu_12 (ReLU)             (None, 2, 2, 512)         0

dropout_12 (Dropout)        (None, 2, 2, 512)         0

max_pooling2d_4 (MaxPooling  (None, 1, 1, 512)        0
2D)

flatten (Flatten)           (None, 512)               0

dense (Dense)               (None, 4096)              2101248

batch_normalization_13 (Bat  (None, 4096)             16384
chNormalization)

re_lu_13 (ReLU)             (None, 4096)              0

dropout_13 (Dropout)        (None, 4096)              0

dense_1 (Dense)             (None, 4096)              16781312

batch_normalization_14 (Bat  (None, 4096)             16384
chNormalization)

re_lu_14 (ReLU)             (None, 4096)              0

dropout_14 (Dropout)        (None, 4096)              0

dense_2 (Dense)             (None, 10)                40970

=================================================================
Total params: 33,687,882
Trainable params: 33,663,050
Non-trainable params: 24,832
_________________________________________________________________
Epoch 1/5
1563/1563 [==============================] - 1765s 1s/step - loss: 2.4409 - accuracy: 0.1003
Epoch 2/5
1563/1563 [==============================] - 1831s 1s/step - loss: 2.3026 - accuracy: 0.1000
Epoch 3/5
1563/1563 [==============================] - 1800s 1s/step - loss: 2.3026 - accuracy: 0.1000
Epoch 4/5
1563/1563 [==============================] - 1753s 1s/step - loss: 2.3026 - accuracy: 0.1000
Epoch 5/5
1563/1563 [==============================] - 1644s 1s/step - loss: 2.3026 - accuracy: 0.1000
313/313 [==============================] - 38s 119ms/step - loss: 2.3026 - accuracy: 0.1000

728x90

저작자표시 비영리 변경금지

'즐거운프로그래밍' 카테고리의 다른 글

[딥러닝] 사용자 데이터로 CNN 학습하기- 2. 수집한 데이터 불러오기 (0)	2023.11.06
[딥러닝] 사용자 데이터로 CNN 학습하기- 1. 라벨링하기 (0)	2023.11.06
[딥러닝] 알렉스넷 바탕으로 신경망 구성하기, 배치 정규화, 드롭아웃 적용하기 (0)	2023.11.02
[딥러닝] 딥러닝 모델 학습 후 저장하고 필터 이미지 추출하기 (1)	2023.11.02
[딥러닝] CNN 활용 예제(엔비디아 자료 보고 신경망 구성하기) (1)	2023.11.01

현재글[딥러닝] VGG16 바탕으로 신경망 구성하기, 배치 정규화, 드롭아웃 적용하기

250x250

고양이 집사겸 노동자 이런거 저런거 테스트 중

딥러닝, python, 자바기초, 개발, Java, 파이썬, 자바, 자바예제, 경제, 파이썬공부, 주식, 프로그래밍, 자바연습, 파이썬초보, 인공신경망학습, 자바웹, 파이썬기초, 자바코딩, 딥러닝공부, 자바공부,

Today :
Yesterday :

수수께끼 고양이