딥러닝 텐서플로우 이미지 증강 예시

Image Augmentation¶

Cats v Dogs 로 다음처럼 모델링 하고, 학습시켜본다.

4 convolutional layers with 32, 64, 128 and 128 convolutions

train for 100 epochs

데이터 제너레이터를 통해 이미지를 증강한다.¶

train_datagen = ImageDataGenerator(
      rotation_range=40,
      width_shift_range=0.2,
      height_shift_range=0.2,
      shear_range=0.2,
      zoom_range=0.2,
      horizontal_flip=True,
      fill_mode='nearest')

• rotation_range is a value in degrees (0–180), a range within which to randomly rotate pictures.
• width_shift and height_shift are ranges (as a fraction of total width or height) within which to randomly translate pictures vertically or horizontally.
• shear_range is for randomly applying shearing transformations.
• zoom_range is for randomly zooming inside pictures.
• horizontal_flip is for randomly flipping half of the images horizontally. This is relevant when there are no assumptions of horizontal assymmetry (e.g. real-world pictures).
• fill_mode is the strategy used for filling in newly created pixels, which can appear after a rotation or a width/height shift.

In [16]:

In [17]:

!wget --no-check-certificate \
    https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip \
    -O /tmp/cats_and_dogs_filtered.zip
  

--2022-12-30 06:44:42--  https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip
Resolving storage.googleapis.com (storage.googleapis.com)... 108.177.97.128, 108.177.125.128, 74.125.203.128, ...
Connecting to storage.googleapis.com (storage.googleapis.com)|108.177.97.128|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 68606236 (65M) [application/zip]
Saving to: ‘/tmp/cats_and_dogs_filtered.zip’

/tmp/cats_and_dogs_ 100%[===================>]  65.43M  22.4MB/s    in 2.9s    

2022-12-30 06:44:45 (22.4 MB/s) - ‘/tmp/cats_and_dogs_filtered.zip’ saved [68606236/68606236]

In [18]:

import zipfile

In [19]:

file = zipfile.ZipFile('/tmp/cats_and_dogs_filtered.zip')

In [20]:

file.extractall('/tmp')

In [20]:

In [21]:

base_dir = '/tmp/cats_and_dogs_filtered'

In [22]:

train_dir = '/tmp/cats_and_dogs_filtered/train'

In [23]:

test_dir =  '/tmp/cats_and_dogs_filtered/validation'

In [23]:

In [24]:

import tensorflow as tf
from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D, Flatten, Dense

In [38]:

def build_model() :
  model = Sequential()
  model.add( Conv2D(16, (3,3), activation='relu', input_shape=(150, 150, 3) ) )
  model.add( MaxPooling2D((2,2) , 2 )) 
  model.add( Conv2D(32, (3,3), activation='relu' ) )
  model.add( MaxPooling2D((2,2) , 2 ))
  model.add( Conv2D(64, (3,3), activation='relu' ) )
  model.add( MaxPooling2D((2,2) , 2 ))

  model.add(Flatten())
  model.add(Dense(512, 'relu'))
  model.add(Dense(1, 'sigmoid'))

  model.compile('rmsprop', 'binary_crossentropy', metrics=['accuracy'])
  return model

In [39]:

model = build_model()

In [40]:

# 이미지 데이터를 증강하여, 학습을 시키자
# 몇배로 증강하는지는 메뉴얼을 참조

In [41]:

from keras.preprocessing.image import ImageDataGenerator

In [42]:

train_datagen = ImageDataGenerator(rescale=1/255.0, rotation_range=30, 
                                   width_shift_range=0.4, 
                                   height_shift_range=0.2,
                                   shear_range=0.3,
                                   zoom_range=0.5,
                                   horizontal_flip=True)

In [43]:

test_datagen = ImageDataGenerator(rescale=1/255.0 )

In [44]:

train_generator = train_datagen.flow_from_directory(train_dir, (150,150), class_mode = 'binary', batch_size=20)  

Found 2000 images belonging to 2 classes.

In [45]:

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

Found 1000 images belonging to 2 classes.

In [46]:

epoch_history = model.fit(train_generator, epochs= 15, validation_data= (test_generator) , steps_per_epoch= 100 ) # 실무에서는 콜백이 반드시 들어간다. 다한다고 최상의 조건이 나오는것은 아니기 때문 (얼리스탑,체크포인트 등 사용)

Epoch 1/15
100/100 [==============================] - 77s 756ms/step - loss: 0.7559 - accuracy: 0.5180 - val_loss: 0.6784 - val_accuracy: 0.5880
Epoch 2/15
100/100 [==============================] - 72s 716ms/step - loss: 0.6881 - accuracy: 0.5555 - val_loss: 0.6512 - val_accuracy: 0.5720
Epoch 3/15
100/100 [==============================] - 72s 719ms/step - loss: 0.6756 - accuracy: 0.5925 - val_loss: 0.6527 - val_accuracy: 0.6260
Epoch 4/15
100/100 [==============================] - 75s 744ms/step - loss: 0.6736 - accuracy: 0.6230 - val_loss: 0.6042 - val_accuracy: 0.6490
Epoch 5/15
100/100 [==============================] - 72s 719ms/step - loss: 0.6522 - accuracy: 0.6140 - val_loss: 0.6599 - val_accuracy: 0.5830
Epoch 6/15
100/100 [==============================] - 72s 719ms/step - loss: 0.6559 - accuracy: 0.6425 - val_loss: 0.5974 - val_accuracy: 0.6610
Epoch 7/15
100/100 [==============================] - 74s 744ms/step - loss: 0.6281 - accuracy: 0.6490 - val_loss: 0.6170 - val_accuracy: 0.6310
Epoch 8/15
100/100 [==============================] - 72s 716ms/step - loss: 0.6320 - accuracy: 0.6415 - val_loss: 0.5858 - val_accuracy: 0.6960
Epoch 9/15
100/100 [==============================] - 72s 718ms/step - loss: 0.6230 - accuracy: 0.6425 - val_loss: 0.8048 - val_accuracy: 0.6220
Epoch 10/15
100/100 [==============================] - 74s 740ms/step - loss: 0.6271 - accuracy: 0.6395 - val_loss: 0.6399 - val_accuracy: 0.6690
Epoch 11/15
100/100 [==============================] - 71s 713ms/step - loss: 0.6095 - accuracy: 0.6660 - val_loss: 0.6034 - val_accuracy: 0.6720
Epoch 12/15
100/100 [==============================] - 71s 712ms/step - loss: 0.6115 - accuracy: 0.6585 - val_loss: 0.7509 - val_accuracy: 0.6440
Epoch 13/15
100/100 [==============================] - 74s 741ms/step - loss: 0.6138 - accuracy: 0.6660 - val_loss: 0.6570 - val_accuracy: 0.6490
Epoch 14/15
100/100 [==============================] - 71s 711ms/step - loss: 0.6047 - accuracy: 0.6695 - val_loss: 0.5661 - val_accuracy: 0.7070
Epoch 15/15
100/100 [==============================] - 74s 735ms/step - loss: 0.6031 - accuracy: 0.6730 - val_loss: 0.5531 - val_accuracy: 0.7150

In [33]: