데이터 증식을 사용한 특성 추출

• 강아지와 고양이 데이터를 캐글로 부터 데이터를 준비
• 사전 훈련 모델에 대해 알아봅니다.
• VGG16 모델을 불러와 사용해 봅니다.
  • 데이터 증식을 사용함.
• FCL 분류기를 연결하여 모델 학습
• 일부 층을 훈련하여 성능을 확인해 봅니다.

import os, shutil

### 드라이브 마운트
from google.colab import drive
drive.mount('/content/drive')

Mounted at /content/drive

#!cp -r '/content/drive/My Drive/dataset/cats_dogs'  '/content/'
#!ls -ls '/content/cats_dogs'

# 압축풀기
#!rm -rf '/content/datasets/'
#!unzip '/content/cats_dogs/test1.zip' -d '/content/datasets/' 
#!unzip '/content/cats_dogs/train.zip' -d '/content/datasets/' 

!ls -al '/content/datasets/train' | head -5
!ls -l '/content/datasets/train' | grep ^- | wc -l
!ls -al '/content/datasets/test1' | head -5
!ls -l '/content/datasets/test1' | grep ^- | wc -l

total 609256
drwxr-xr-x 2 root root 765952 Sep 20  2013 .
drwxr-xr-x 4 root root   4096 Nov 10 16:19 ..
-rw-r--r-- 1 root root  12414 Sep 20  2013 cat.0.jpg
-rw-r--r-- 1 root root  21944 Sep 20  2013 cat.10000.jpg
25000
total 304280
drwxr-xr-x 2 root root 290816 Sep 20  2013 .
drwxr-xr-x 4 root root   4096 Nov 10 16:19 ..
-rw-r--r-- 1 root root  54902 Sep 20  2013 10000.jpg
-rw-r--r-- 1 root root  21671 Sep 20  2013 10001.jpg
12500

데이터 준비

# 원본 데이터셋을 압축 해제한 디렉터리 경로
ori_dataset_dir = './datasets/train'

# 소규모 데이터셋을 저장할 디렉터리
base_dir = './datasets/cats_and_dogs_small'

# 반복실행을 위해 디렉터리 삭제
if os.path.exists(base_dir):
    shutil.rmtree(base_dir)
os.mkdir(base_dir)

# 훈련, 검증, 테스트 분할을 위한 디렉터리
train_dir = os.path.join(base_dir, 'train')
os.mkdir(train_dir)

val_dir = os.path.join(base_dir, 'validation')
os.mkdir(val_dir)

test_dir = os.path.join(base_dir, 'test')
os.mkdir(test_dir)

# 훈련용 고양이 사진 디렉터리
train_cats_dir = os.path.join(train_dir, 'cats')
os.mkdir(train_cats_dir)

# 훈련용 강아지 사진 디렉터리
train_dogs_dir = os.path.join(train_dir, 'dogs')
os.mkdir(train_dogs_dir)

# 검증용 고양이 사진 디렉터리
val_cats_dir = os.path.join(val_dir, 'cats')
os.mkdir(val_cats_dir)

# 검증용 강아지 사진 디렉터리
val_dogs_dir = os.path.join(val_dir, 'dogs')
os.mkdir(val_dogs_dir)

# 테스트용 고양이 사진 디렉터리
test_cats_dir = os.path.join(test_dir, 'cats')
os.mkdir(test_cats_dir)

# 테스트용 강아지 사진 디렉터리
test_dogs_dir = os.path.join(test_dir, 'dogs')
os.mkdir(test_dogs_dir)

# 처음 1,000개의 고양이 이미지를 train_cats_dir에 복사합니다
fnames = ['cat.{}.jpg'.format(i) for i in range(1000)]
for fname in fnames:
    src = os.path.join(ori_dataset_dir, fname)
    dst = os.path.join(train_cats_dir, fname)
    shutil.copyfile(src, dst)

# 다음 500개 고양이 이미지를 validation_cats_dir에 복사합니다
fnames = ['cat.{}.jpg'.format(i) for i in range(1000, 1500)]
for fname in fnames:
    src = os.path.join(ori_dataset_dir, fname)
    dst = os.path.join(val_cats_dir, fname)
    shutil.copyfile(src, dst)
    
# 다음 500개 고양이 이미지를 test_cats_dir에 복사합니다
fnames = ['cat.{}.jpg'.format(i) for i in range(1500, 2000)]
for fname in fnames:
    src = os.path.join(ori_dataset_dir, fname)
    dst = os.path.join(test_cats_dir, fname)
    shutil.copyfile(src, dst)
    
# 처음 1,000개의 강아지 이미지를 train_dogs_dir에 복사합니다
fnames = ['dog.{}.jpg'.format(i) for i in range(1000)]
for fname in fnames:
    src = os.path.join(ori_dataset_dir, fname)
    dst = os.path.join(train_dogs_dir, fname)
    shutil.copyfile(src, dst)
    
# 다음 500개 강아지 이미지를 validation_dogs_dir에 복사합니다
fnames = ['dog.{}.jpg'.format(i) for i in range(1000, 1500)]
for fname in fnames:
    src = os.path.join(ori_dataset_dir, fname)
    dst = os.path.join(val_dogs_dir, fname)
    shutil.copyfile(src, dst)
    
# 다음 500개 강아지 이미지를 test_dogs_dir에 복사합니다
fnames = ['dog.{}.jpg'.format(i) for i in range(1500, 2000)]
for fname in fnames:
    src = os.path.join(ori_dataset_dir, fname)
    dst = os.path.join(test_dogs_dir, fname)
    shutil.copyfile(src, dst)

print('훈련용 고양이 이미지 전체 개수:', len(os.listdir(train_cats_dir)))
print('훈련용 강아지 이미지 전체 개수:', len(os.listdir(train_dogs_dir)))

print('검증용 고양이 이미지 전체 개수:', len(os.listdir(val_cats_dir)))
print('검증용 강아지 이미지 전체 개수:', len(os.listdir(val_dogs_dir)))

print('테스트용 고양이 이미지 전체 개수:', len(os.listdir(test_cats_dir)))
print('테스트용 강아지 이미지 전체 개수:', len(os.listdir(test_dogs_dir)))

훈련용 고양이 이미지 전체 개수: 1000
훈련용 강아지 이미지 전체 개수: 1000
검증용 고양이 이미지 전체 개수: 500
검증용 강아지 이미지 전체 개수: 500
테스트용 고양이 이미지 전체 개수: 500
테스트용 강아지 이미지 전체 개수: 500

## 경로에 이미지 데이터의 개수 
num_cats_tr = len(os.listdir(train_cats_dir))
num_dogs_tr = len(os.listdir(train_dogs_dir))

num_cats_val = len(os.listdir(val_cats_dir))
num_dogs_val = len(os.listdir(val_dogs_dir))

total_train = num_cats_tr + num_dogs_tr
total_val = num_cats_val + num_dogs_val

print("학습용 데이터 : ", total_train)
print("검증용 데이터 : ", total_val)

학습용 데이터 :  2000
검증용 데이터 :  1000

모델 구축

from tensorflow.keras.applications import VGG16

conv_base = VGG16(weights='imagenet',
                  include_top=False,
                  input_shape=(150, 150, 3))

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/vgg16/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5
58892288/58889256 [==============================] - 1s 0us/step
58900480/58889256 [==============================] - 1s 0us/step

from tensorflow.keras import models
from tensorflow.keras import layers

model = models.Sequential()
model.add(conv_base)
model.add(layers.Flatten())
model.add(layers.Dense(256, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))

model.summary()

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 vgg16 (Functional)          (None, 4, 4, 512)         14714688  
                                                                 
 flatten (Flatten)           (None, 8192)              0         
                                                                 
 dense (Dense)               (None, 256)               2097408   
                                                                 
 dense_1 (Dense)             (None, 1)                 257       
                                                                 
=================================================================
Total params: 16,812,353
Trainable params: 16,812,353
Non-trainable params: 0
_________________________________________________________________

• VGG16의 합성곱 기반층은 14,714,688개의 매우 많은 파라미터 보유
• 모델을 컴파일하고 훈련하기 전에 합성곱 기반층을 동결하는 것이 아주 중요
  • 하나 이상의 층을 동결한다는 것은 훈련하는 동안 가중치가 업데이트되지 않도록 막는다는 뜻
  • 동결하지 않으면 합성곱 기반층에 의해 사전에 학습된 표현이 훈련하는 동안 수정
  • 맨 위의 Dense 층은 랜덤하게 초기화되었기 때문에 매우 큰 가중치 업데이트 값이 네트워크에 전파
  • 케라스에서는 trainable 속성을 False로 설정하여 네트워크를 동결

print('conv_base를 동결 전 훈련되는 가중치의 수:', len(model.trainable_weights))
conv_base.trainable = False
print('conv_base를 동결 후 훈련되는 가중치의 수:',   len(model.trainable_weights))

conv_base를 동결 전 훈련되는 가중치의 수: 30
conv_base를 동결 후 훈련되는 가중치의 수: 4

• 2개의 Dense층 가중치만 훈련
• 각 층마다 2개씩(가중치 행렬과 편향 벡터) 총 4개의 가중치(텐서)가 훈련
• 컴파일 단계 후에 trainable 속성을 변경하면 반드시 모델을 다시 컴파일 해야 한다.

from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras import optimizers

batch_size = 20
epochs = 30

• rotation_range : 랜덤하게 사진을 회전시킬 각도 범위(0~180 사이)
• width_shift_range와 height_shift_range는 사진을 수평과 수직으로 랜덤하게 평행 이동 시킬 범위
• shear_range : 랜덤하게 전단 변환(shearing transformation)을 적용할 각도 범위
• zoom_range : 랜덤하게 사진을 확대할 범위
• horizontal_flip : 랜덤하게 이미지를 수평으로 뒤집기. 수평 대칭을 가정할 수 있을 때 사용.(예 풍경/인물 사진)
• fill_mode : 회전이나 가로/세로 이동으로 인해 새롭게 생성해야 할 픽셀을 채울 전략

train_datagen = ImageDataGenerator(
      rescale=1./255,
      rotation_range=20,
      width_shift_range=0.1,
      height_shift_range=0.1,
      shear_range=0.1,
      zoom_range=0.1,
      horizontal_flip=True,
      fill_mode='nearest')

train_generator = train_datagen.flow_from_directory(
        # 타깃 디렉터리
        train_dir,
        # 모든 이미지의 크기를 150 × 150로 변경합니다
        target_size=(150, 150),
        batch_size=20,
        # binary_crossentropy 손실을 사용하므로 이진 레이블이 필요합니다
        class_mode='binary')

# 검증 데이터는 증식되어서는 안 됩니다!
test_datagen = ImageDataGenerator(rescale=1./255)

validation_generator = test_datagen.flow_from_directory(
        val_dir,
        target_size=(150, 150),
        batch_size=20,
        class_mode='binary')

Found 2000 images belonging to 2 classes.
Found 1000 images belonging to 2 classes.

%%time

model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=2e-5),
              metrics=['acc'])

history = model.fit_generator(
            train_generator,
            steps_per_epoch=100,  # total_train // batch_size
            epochs=30,
            validation_data=validation_generator,
            validation_steps=50,  # total_val // batch_size
            verbose=2)

/usr/local/lib/python3.7/dist-packages/keras/optimizer_v2/rmsprop.py:130: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead.
  super(RMSprop, self).__init__(name, **kwargs)
/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:11: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.
  # This is added back by InteractiveShellApp.init_path()

Epoch 1/30
100/100 - 59s - loss: 0.5314 - acc: 0.7560 - val_loss: 0.4081 - val_acc: 0.8210 - 59s/epoch - 589ms/step
Epoch 2/30
100/100 - 26s - loss: 0.4006 - acc: 0.8345 - val_loss: 0.3440 - val_acc: 0.8540 - 26s/epoch - 259ms/step
Epoch 3/30
100/100 - 26s - loss: 0.3520 - acc: 0.8525 - val_loss: 0.3028 - val_acc: 0.8770 - 26s/epoch - 256ms/step
Epoch 4/30
100/100 - 25s - loss: 0.3211 - acc: 0.8710 - val_loss: 0.2839 - val_acc: 0.8880 - 25s/epoch - 254ms/step
Epoch 5/30
100/100 - 26s - loss: 0.3059 - acc: 0.8660 - val_loss: 0.2729 - val_acc: 0.8990 - 26s/epoch - 256ms/step
Epoch 6/30
100/100 - 26s - loss: 0.2930 - acc: 0.8780 - val_loss: 0.2674 - val_acc: 0.8930 - 26s/epoch - 256ms/step
Epoch 7/30
100/100 - 26s - loss: 0.2797 - acc: 0.8845 - val_loss: 0.2583 - val_acc: 0.8980 - 26s/epoch - 257ms/step
Epoch 8/30
100/100 - 25s - loss: 0.2728 - acc: 0.8925 - val_loss: 0.2597 - val_acc: 0.8940 - 25s/epoch - 255ms/step
Epoch 9/30
100/100 - 26s - loss: 0.2615 - acc: 0.8965 - val_loss: 0.2524 - val_acc: 0.8980 - 26s/epoch - 256ms/step
Epoch 10/30
100/100 - 25s - loss: 0.2495 - acc: 0.8930 - val_loss: 0.2481 - val_acc: 0.9000 - 25s/epoch - 255ms/step
Epoch 11/30
100/100 - 25s - loss: 0.2499 - acc: 0.9035 - val_loss: 0.2455 - val_acc: 0.8980 - 25s/epoch - 255ms/step
Epoch 12/30
100/100 - 25s - loss: 0.2408 - acc: 0.9020 - val_loss: 0.2444 - val_acc: 0.8980 - 25s/epoch - 252ms/step
Epoch 13/30
100/100 - 25s - loss: 0.2380 - acc: 0.9070 - val_loss: 0.2444 - val_acc: 0.9040 - 25s/epoch - 253ms/step
Epoch 14/30
100/100 - 25s - loss: 0.2218 - acc: 0.9115 - val_loss: 0.2512 - val_acc: 0.8970 - 25s/epoch - 251ms/step
Epoch 15/30
100/100 - 25s - loss: 0.2217 - acc: 0.9090 - val_loss: 0.2392 - val_acc: 0.9040 - 25s/epoch - 253ms/step
Epoch 16/30
100/100 - 25s - loss: 0.2218 - acc: 0.9125 - val_loss: 0.2459 - val_acc: 0.9010 - 25s/epoch - 254ms/step
Epoch 17/30
100/100 - 26s - loss: 0.2241 - acc: 0.9080 - val_loss: 0.2409 - val_acc: 0.9000 - 26s/epoch - 256ms/step
Epoch 18/30
100/100 - 25s - loss: 0.2164 - acc: 0.9105 - val_loss: 0.2366 - val_acc: 0.9020 - 25s/epoch - 253ms/step
Epoch 19/30
100/100 - 25s - loss: 0.2038 - acc: 0.9200 - val_loss: 0.2511 - val_acc: 0.8950 - 25s/epoch - 254ms/step
Epoch 20/30
100/100 - 25s - loss: 0.2090 - acc: 0.9195 - val_loss: 0.2372 - val_acc: 0.9050 - 25s/epoch - 254ms/step
Epoch 21/30
100/100 - 25s - loss: 0.2012 - acc: 0.9225 - val_loss: 0.2363 - val_acc: 0.9020 - 25s/epoch - 251ms/step
Epoch 22/30
100/100 - 25s - loss: 0.2096 - acc: 0.9140 - val_loss: 0.2360 - val_acc: 0.8990 - 25s/epoch - 251ms/step
Epoch 23/30
100/100 - 25s - loss: 0.2083 - acc: 0.9245 - val_loss: 0.2351 - val_acc: 0.9030 - 25s/epoch - 252ms/step
Epoch 24/30
100/100 - 25s - loss: 0.2057 - acc: 0.9175 - val_loss: 0.2412 - val_acc: 0.9000 - 25s/epoch - 254ms/step
Epoch 25/30
100/100 - 25s - loss: 0.1910 - acc: 0.9290 - val_loss: 0.2473 - val_acc: 0.9010 - 25s/epoch - 251ms/step
Epoch 26/30
100/100 - 25s - loss: 0.1928 - acc: 0.9220 - val_loss: 0.2353 - val_acc: 0.9080 - 25s/epoch - 250ms/step
Epoch 27/30
100/100 - 25s - loss: 0.1919 - acc: 0.9230 - val_loss: 0.2391 - val_acc: 0.9080 - 25s/epoch - 250ms/step
Epoch 28/30
100/100 - 25s - loss: 0.1856 - acc: 0.9275 - val_loss: 0.2402 - val_acc: 0.9020 - 25s/epoch - 251ms/step
Epoch 29/30
100/100 - 25s - loss: 0.1815 - acc: 0.9300 - val_loss: 0.2395 - val_acc: 0.9030 - 25s/epoch - 251ms/step
Epoch 30/30
100/100 - 25s - loss: 0.1859 - acc: 0.9280 - val_loss: 0.2393 - val_acc: 0.9030 - 25s/epoch - 251ms/step

모델 저장

model.save('cats_dogs_small_3_dataAug_VGG_30_epoch.h5')

결과 시각화

import matplotlib.pyplot as plt

acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']

epochs = range(len(acc))

plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()

plt.figure()

plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()

plt.show()

미세 조정 - 일부 훈련

• 모델을 재사용하는 데 널리 사용되는 또 하나의 기법은 특성 추출을 보완하는 미세 조정
• 미세 조정은 특성 추출에 사용했던 동결 모델의 상위 층 몇 개를 동결에서 해제한다.
  • 모델에 새로 추가한 층(여기서는 완전 연결 분류기)과 함께 훈련하는 것.
  • 주어진 문제에 조금 더 밀접하게 재사용 모델의 표현을 일부 조정하기 때문에 미세 조정이라고 부른다.

• 같은 이유로 맨 위에 있는 분류기가 훈련된 후에 합성곱 기반의 상위 층을 미세 조정 가능.

• 분류기가 미리 훈련되지 않으면 훈련되는 동안 너무 큰 오차 신호가 네트워크에 전파됩니다.

  • 미세 조정될 층들이 사전에 학습한 표현들을 망가뜨리게 될 것입니다.
  • 네트워크를 미세 조정하는 단계:
    • 01 사전에 훈련된 기반 네트워크 위에 새로운 네트워크를 추가합니다.
    • 02 훈련된 기반 네트워크를 동결합니다.
    • 03 새로 추가한 네트워크를 훈련합니다.
    • 04 기반 네트워크에서 일부 층의 동결을 해제합니다.
    • 05 동결을 해제한 층과 새로 추가한 층을 함께 훈련합니다.

• 처음 세 단계는 특성 추출을 할 때 이미 완료했습니다.

• 네 번째 단계를 진행해 보죠. conv_base의 동결을 해제하고 개별 층을 동결하겠습니다.

conv_base.summary()

Model: "vgg16"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_1 (InputLayer)        [(None, 150, 150, 3)]     0         
                                                                 
 block1_conv1 (Conv2D)       (None, 150, 150, 64)      1792      
                                                                 
 block1_conv2 (Conv2D)       (None, 150, 150, 64)      36928     
                                                                 
 block1_pool (MaxPooling2D)  (None, 75, 75, 64)        0         
                                                                 
 block2_conv1 (Conv2D)       (None, 75, 75, 128)       73856     
                                                                 
 block2_conv2 (Conv2D)       (None, 75, 75, 128)       147584    
                                                                 
 block2_pool (MaxPooling2D)  (None, 37, 37, 128)       0         
                                                                 
 block3_conv1 (Conv2D)       (None, 37, 37, 256)       295168    
                                                                 
 block3_conv2 (Conv2D)       (None, 37, 37, 256)       590080    
                                                                 
 block3_conv3 (Conv2D)       (None, 37, 37, 256)       590080    
                                                                 
 block3_pool (MaxPooling2D)  (None, 18, 18, 256)       0         
                                                                 
 block4_conv1 (Conv2D)       (None, 18, 18, 512)       1180160   
                                                                 
 block4_conv2 (Conv2D)       (None, 18, 18, 512)       2359808   
                                                                 
 block4_conv3 (Conv2D)       (None, 18, 18, 512)       2359808   
                                                                 
 block4_pool (MaxPooling2D)  (None, 9, 9, 512)         0         
                                                                 
 block5_conv1 (Conv2D)       (None, 9, 9, 512)         2359808   
                                                                 
 block5_conv2 (Conv2D)       (None, 9, 9, 512)         2359808   
                                                                 
 block5_conv3 (Conv2D)       (None, 9, 9, 512)         2359808   
                                                                 
 block5_pool (MaxPooling2D)  (None, 4, 4, 512)         0         
                                                                 
=================================================================
Total params: 14,714,688
Trainable params: 0
Non-trainable params: 14,714,688
_________________________________________________________________

• 마지막 세 개의 합성곱 층을 미세 조정하겠습니다.
  • 즉, block4_pool까지 모든 층은 동결
  • block5_conv1, block5_conv2, block5_conv3 층은 학습 대상

conv_base.trainable = True

set_trainable = False
for layer in conv_base.layers:
    if layer.name == 'block5_conv1':
        set_trainable = True
    if set_trainable:
        layer.trainable = True
    else:
        layer.trainable = False

%%time

model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=1e-5),
              metrics=['acc'])


history = model.fit_generator(
      train_generator,
      steps_per_epoch=100,  # total_train // batch_size
      epochs=50,
      validation_data=validation_generator,
      validation_steps=50,  # total_val // batch_size
      verbose=1)

/usr/local/lib/python3.7/dist-packages/keras/optimizer_v2/rmsprop.py:130: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead.
  super(RMSprop, self).__init__(name, **kwargs)
/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:13: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.
  del sys.path[0]

Epoch 1/50
100/100 [==============================] - 30s 271ms/step - loss: 0.2049 - acc: 0.9125 - val_loss: 0.2927 - val_acc: 0.8840
Epoch 2/50
100/100 [==============================] - 27s 268ms/step - loss: 0.1486 - acc: 0.9435 - val_loss: 0.2194 - val_acc: 0.9220
Epoch 3/50
100/100 [==============================] - 27s 269ms/step - loss: 0.1357 - acc: 0.9475 - val_loss: 0.2155 - val_acc: 0.9170
Epoch 4/50
100/100 [==============================] - 27s 267ms/step - loss: 0.1167 - acc: 0.9605 - val_loss: 0.2070 - val_acc: 0.9190
Epoch 5/50
100/100 [==============================] - 27s 270ms/step - loss: 0.1085 - acc: 0.9580 - val_loss: 0.2113 - val_acc: 0.9200
Epoch 6/50
100/100 [==============================] - 27s 269ms/step - loss: 0.0840 - acc: 0.9675 - val_loss: 0.2349 - val_acc: 0.9110
Epoch 7/50
100/100 [==============================] - 27s 271ms/step - loss: 0.0924 - acc: 0.9680 - val_loss: 0.2123 - val_acc: 0.9200
Epoch 8/50
100/100 [==============================] - 27s 267ms/step - loss: 0.0712 - acc: 0.9740 - val_loss: 0.2333 - val_acc: 0.9210
Epoch 9/50
100/100 [==============================] - 27s 268ms/step - loss: 0.0582 - acc: 0.9780 - val_loss: 0.2257 - val_acc: 0.9280
Epoch 10/50
100/100 [==============================] - 27s 270ms/step - loss: 0.0519 - acc: 0.9830 - val_loss: 0.2757 - val_acc: 0.9120
Epoch 11/50
100/100 [==============================] - 27s 272ms/step - loss: 0.0546 - acc: 0.9830 - val_loss: 0.2997 - val_acc: 0.9070
Epoch 12/50
100/100 [==============================] - 27s 273ms/step - loss: 0.0501 - acc: 0.9850 - val_loss: 0.2472 - val_acc: 0.9150
Epoch 13/50
100/100 [==============================] - 27s 271ms/step - loss: 0.0501 - acc: 0.9840 - val_loss: 0.2334 - val_acc: 0.9210
Epoch 14/50
100/100 [==============================] - 27s 271ms/step - loss: 0.0458 - acc: 0.9855 - val_loss: 0.2115 - val_acc: 0.9210
Epoch 15/50
100/100 [==============================] - 27s 270ms/step - loss: 0.0335 - acc: 0.9875 - val_loss: 0.2254 - val_acc: 0.9240
Epoch 16/50
100/100 [==============================] - 27s 270ms/step - loss: 0.0286 - acc: 0.9895 - val_loss: 0.2243 - val_acc: 0.9300
Epoch 17/50
100/100 [==============================] - 27s 270ms/step - loss: 0.0294 - acc: 0.9940 - val_loss: 0.2262 - val_acc: 0.9270
Epoch 18/50
100/100 [==============================] - 27s 275ms/step - loss: 0.0269 - acc: 0.9900 - val_loss: 0.2338 - val_acc: 0.9240
Epoch 19/50
100/100 [==============================] - 28s 275ms/step - loss: 0.0246 - acc: 0.9905 - val_loss: 0.2649 - val_acc: 0.9150
Epoch 20/50
100/100 [==============================] - 27s 269ms/step - loss: 0.0237 - acc: 0.9925 - val_loss: 0.3777 - val_acc: 0.9030
Epoch 21/50
100/100 [==============================] - 27s 267ms/step - loss: 0.0184 - acc: 0.9935 - val_loss: 0.3034 - val_acc: 0.9130
Epoch 22/50
100/100 [==============================] - 26s 264ms/step - loss: 0.0221 - acc: 0.9925 - val_loss: 0.2562 - val_acc: 0.9250
Epoch 23/50
100/100 [==============================] - 26s 264ms/step - loss: 0.0185 - acc: 0.9940 - val_loss: 0.2645 - val_acc: 0.9210
Epoch 24/50
100/100 [==============================] - 27s 265ms/step - loss: 0.0216 - acc: 0.9920 - val_loss: 0.2532 - val_acc: 0.9260
Epoch 25/50
100/100 [==============================] - 26s 264ms/step - loss: 0.0104 - acc: 0.9975 - val_loss: 0.2571 - val_acc: 0.9280
Epoch 26/50
100/100 [==============================] - 27s 266ms/step - loss: 0.0145 - acc: 0.9955 - val_loss: 0.4208 - val_acc: 0.9070
Epoch 27/50
100/100 [==============================] - 27s 267ms/step - loss: 0.0147 - acc: 0.9960 - val_loss: 0.3010 - val_acc: 0.9260
Epoch 28/50
100/100 [==============================] - 27s 270ms/step - loss: 0.0131 - acc: 0.9975 - val_loss: 0.3044 - val_acc: 0.9230
Epoch 29/50
100/100 [==============================] - 27s 269ms/step - loss: 0.0113 - acc: 0.9955 - val_loss: 0.3494 - val_acc: 0.9260
Epoch 30/50
100/100 [==============================] - 27s 268ms/step - loss: 0.0119 - acc: 0.9950 - val_loss: 0.2851 - val_acc: 0.9300
Epoch 31/50
100/100 [==============================] - 27s 268ms/step - loss: 0.0117 - acc: 0.9955 - val_loss: 0.2791 - val_acc: 0.9300
Epoch 32/50
100/100 [==============================] - 27s 267ms/step - loss: 0.0168 - acc: 0.9950 - val_loss: 0.2667 - val_acc: 0.9270
Epoch 33/50
100/100 [==============================] - 27s 267ms/step - loss: 0.0070 - acc: 0.9980 - val_loss: 0.3492 - val_acc: 0.9250
Epoch 34/50
100/100 [==============================] - 27s 266ms/step - loss: 0.0106 - acc: 0.9960 - val_loss: 0.3605 - val_acc: 0.9200
Epoch 35/50
100/100 [==============================] - 27s 269ms/step - loss: 0.0135 - acc: 0.9950 - val_loss: 0.3302 - val_acc: 0.9180
Epoch 36/50
100/100 [==============================] - 27s 268ms/step - loss: 0.0054 - acc: 0.9975 - val_loss: 0.3356 - val_acc: 0.9280
Epoch 37/50
100/100 [==============================] - 27s 267ms/step - loss: 0.0051 - acc: 0.9985 - val_loss: 0.2848 - val_acc: 0.9330
Epoch 38/50
100/100 [==============================] - 27s 267ms/step - loss: 0.0121 - acc: 0.9950 - val_loss: 0.3051 - val_acc: 0.9260
Epoch 39/50
100/100 [==============================] - 26s 263ms/step - loss: 0.0087 - acc: 0.9965 - val_loss: 0.3200 - val_acc: 0.9230
Epoch 40/50
100/100 [==============================] - 27s 266ms/step - loss: 0.0067 - acc: 0.9980 - val_loss: 0.3095 - val_acc: 0.9260
Epoch 41/50
100/100 [==============================] - 27s 265ms/step - loss: 0.0031 - acc: 0.9995 - val_loss: 0.3152 - val_acc: 0.9280
Epoch 42/50
100/100 [==============================] - 27s 265ms/step - loss: 0.0128 - acc: 0.9965 - val_loss: 0.2791 - val_acc: 0.9280
Epoch 43/50
100/100 [==============================] - 26s 264ms/step - loss: 0.0055 - acc: 0.9985 - val_loss: 0.3054 - val_acc: 0.9360
Epoch 44/50
100/100 [==============================] - 27s 265ms/step - loss: 0.0061 - acc: 0.9990 - val_loss: 0.2962 - val_acc: 0.9300
Epoch 45/50
100/100 [==============================] - 27s 266ms/step - loss: 0.0129 - acc: 0.9965 - val_loss: 0.2914 - val_acc: 0.9330
Epoch 46/50
100/100 [==============================] - 27s 266ms/step - loss: 0.0061 - acc: 0.9975 - val_loss: 0.4213 - val_acc: 0.9120
Epoch 47/50
100/100 [==============================] - 27s 265ms/step - loss: 0.0045 - acc: 0.9975 - val_loss: 0.3242 - val_acc: 0.9350
Epoch 48/50
100/100 [==============================] - 26s 264ms/step - loss: 0.0038 - acc: 0.9990 - val_loss: 0.3308 - val_acc: 0.9300
Epoch 49/50
100/100 [==============================] - 26s 264ms/step - loss: 0.0109 - acc: 0.9965 - val_loss: 0.2993 - val_acc: 0.9340
Epoch 50/50
100/100 [==============================] - 26s 263ms/step - loss: 0.0026 - acc: 0.9990 - val_loss: 0.3268 - val_acc: 0.9350
CPU times: user 25min 5s, sys: 1min 6s, total: 26min 11s
Wall time: 22min 52s

## 모델 저장
model.save('cats_and_dogs_small_4_misetunning_50.h5')

시각화

acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']

epochs = range(len(acc))

plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()

plt.figure()

plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()

plt.show()

모델 평가

test_generator = test_datagen.flow_from_directory(
        test_dir,
        target_size=(150, 150),
        batch_size=20,
        class_mode='binary')

test_loss, test_acc = model.evaluate_generator(test_generator, steps=50)
print('test acc:', test_acc)

Found 1000 images belonging to 2 classes.

/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:7: UserWarning: `Model.evaluate_generator` is deprecated and will be removed in a future version. Please use `Model.evaluate`, which supports generators.
  import sys

test acc: 0.9359999895095825