인공지능 기초 🚶‍♂️: 사진 분류 프로그램 맛보기

강아지 & 고양이 분류 프로그램 제작 !

Content

데이터 불러오기 (Tensor Flow)
데이터 전처리
모델 생성 (자체 생성 모델)
학습 및 평가

시작하기에 앞서, 이번 게시글은
전체적인 인공지능 적용에 대한 맥락을 설명하기 위한 글입니다.
아주 간단한 Deep Learning Layer 몇 종류를 이용했습니다.

import tensorflow as tf
print(tf.__version__ , '이미지 분류 모델을 만들 라이브러리 tensor flow 입니다')

2.6.0 이미지 분류 모델을 만들 라이브러리 tensor flow 입니다

import tensorflow_datasets as tfds

tfds.__version__

'4.4.0'

tensor flow 에서는 다양한 데이터셋을 이미 제공하고 있다.
강아지고양이, 음성, 이미지, 텍스트 데이터셋 보유하고 있으니 세부 내용 확인해보고 싶음 해보기

tensor flow link

(raw_train, raw_validation, raw_test), metadata = tfds.load(
    'cats_vs_dogs',
    split=['train[:80%]', 'train[80%:90%]', 'train[90%:]'],
    with_info=True,
    as_supervised=True,
)

[1mDownloading and preparing dataset 786.68 MiB (download: 786.68 MiB, generated: Unknown size, total: 786.68 MiB) to /aiffel/tensorflow_datasets/cats_vs_dogs/4.0.0...[0m



Dl Completed...: 0 url [00:00, ? url/s]



Dl Size...: 0 MiB [00:00, ? MiB/s]



Generating splits...:   0%|          | 0/1 [00:00<?, ? splits/s]



Generating train examples...:   0%|          | 0/23262 [00:00<?, ? examples/s]


WARNING:absl:1738 images were corrupted and were skipped



Shuffling cats_vs_dogs-train.tfrecord...:   0%|          | 0/23262 [00:00<?, ? examples/s]


[1mDataset cats_vs_dogs downloaded and prepared to /aiffel/tensorflow_datasets/cats_vs_dogs/4.0.0. Subsequent calls will reuse this data.[0m

“WARNING:absl:1738 images were corrupted and were skipped”라는 경고가 나타날 수 있습니다. 우선 무시하시면 됩니다.
1738 장의 사진은 쓸 수 없다는 뜻입니다.
이런 것들은 tf 사이트 데이터셋 설명에서 확인 가능합니다

PNG

1738이 currupted 되어있다 되어있져?
그리고 밑에 Split을 보면 23262 장의 사진이 있음을 확인 가능합니다.

print(raw_train)
print(raw_validation)
print(raw_test)

<PrefetchDataset shapes: ((None, None, 3), ()), types: (tf.uint8, tf.int64)>
<PrefetchDataset shapes: ((None, None, 3), ()), types: (tf.uint8, tf.int64)>
<PrefetchDataset shapes: ((None, None, 3), ()), types: (tf.uint8, tf.int64)>

잘 변수로 지정되어있음을 알 수 있다.

모든 데이터셋은 (image, label)의 형태를 가집니다.
((None, None, 3), ())가 이를 나타내죠.

여기에서 앞에 있는 (None, None, 3)은 image의 shape를,
뒤의 ()는 정답 카테고리인 label의 shape를 의미합니다.

이미지는 (height, width, channel)로 3차원 데이터이기 때문에
(None, None, 3)과 같이 나타났습니다.

이때 height와 width가 None으로 나타난 이유는
모든 사진들의 크기가 제각각이기 때문입니다.
하나의 값으로 나타낼 수 없으니 None 으로 표기됩니다.

1. 데이터 전처리

자 데이터를 불러왔으니 깔끔하게 처리해야지

import matplotlib.pyplot as plt
%matplotlib inline
%config InlineBackend.figure_format = 'retina'

데이터를 어떻게 처리할지 고민하기 위해서 우리가 가져온 데이터를
스리슬쩍 들여다 보자.

plt.figure(figsize=(10, 5)) #전체 칸바스 크기

get_label_name = metadata.features['label'].int2str 

for idx, (image, label) in enumerate(raw_train.take(10)):  # 10개의 데이터를 따로 가져 옵니다.
    plt.subplot(2, 5, idx+1) #10개의 사진을 꺼내보겠음 판 꺼냄
    plt.imshow(image)
    plt.title(f'label {label}: {get_label_name(label)}')
    plt.axis('off')

png

자 봐라, 이미지 크기가 다 제각각이네?

올바른 학습을 위해서는 이미지 사이즈부터 제대로 맞춰줘야 한다.

format_example() 함수로 이미지를 같은 포멧으로 맞춥니다.

IMG_SIZE = 160 # 리사이징할 이미지의 크기

def format_example(image, label):
    image = tf.cast(image, tf.float32)  # image=float(image)같은 타입캐스팅의  텐서플로우 버전입니다.
    image = (image/127.5) - 1 # 픽셀값의 scale 수정
    image = tf.image.resize(image, (IMG_SIZE, IMG_SIZE))
    return image, label

픽셀값의 scale 을 수정햇다는 것은 픽셀값을 정규화햇다는 말과 근사하다.
0~ 255인 픽셀값을 127.5로 나누면 0~ 2가된다. 그걸 1로 뺐으니
-1~1 사이의 값으로 변했다.

train = raw_train.map(format_example)
validation = raw_validation.map(format_example)
test = raw_test.map(format_example)

print(train)
print(validation)
print(test)

<MapDataset shapes: ((160, 160, 3), ()), types: (tf.float32, tf.int64)>
<MapDataset shapes: ((160, 160, 3), ()), types: (tf.float32, tf.int64)>
<MapDataset shapes: ((160, 160, 3), ()), types: (tf.float32, tf.int64)>

map 메서드로 모든 raw_** 의 정보를 format_example 의 함수로 변환시켜주었다.
변환된 후의 내용물은 다음과 같다.

plt.figure(figsize=(10, 5))


get_label_name = metadata.features['label'].int2str

for idx, (image, label) in enumerate(train.take(10)):
    plt.subplot(2, 5, idx+1)
    image = (image + 1) / 2
    plt.imshow(image)
    plt.title(f'label {label}: {get_label_name(label)}')
    plt.axis('off')

png

데이터 전처리 끝~

2 모델 생성 및 학습

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

models 에는 모델 자체를 구축하기 위한 함수가 있고 그 안의 Sequential 함수 안에 여러가지 layer 들이 들어갈 수 있다.
layers 에는 모델의 구성 요소인 여러가지 종류의 layer(층) 함수들을 가지고 있다.

model = Sequential([
    Conv2D(filters=16, kernel_size=3, padding='same', activation='relu', input_shape=(160, 160, 3)),
    MaxPooling2D(),
    Conv2D(filters=32, kernel_size=3, padding='same', activation='relu'),
    MaxPooling2D(),
    Conv2D(filters=64, kernel_size=3, padding='same', activation='relu'),
    MaxPooling2D(),
    Flatten(),
    Dense(units=512, activation='relu'),
    Dense(units=2, activation='softmax')
])

딥러닝에서는 레이어 라는 개념을 자세하게 공부한다.
여기서는

Conv2D
MaxPooling2D
Flatten
Dense

라는 네 레이어를 사용했다.

model.summary()

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_3 (Conv2D)            (None, 160, 160, 16)      448       
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 80, 80, 16)        0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 80, 80, 32)        4640      
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 40, 40, 32)        0         
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 40, 40, 64)        18496     
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 20, 20, 64)        0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 25600)             0         
_________________________________________________________________
dense_2 (Dense)              (None, 512)               13107712  
_________________________________________________________________
dense_3 (Dense)              (None, 2)                 1026      
=================================================================
Total params: 13,132,322
Trainable params: 13,132,322
Non-trainable params: 0
_________________________________________________________________

모델을 만들었으니 학습시켜보자.

learning_rate = 0.0001
model.compile(optimizer=tf.keras.optimizers.RMSprop(lr=learning_rate),
              loss=tf.keras.losses.sparse_categorical_crossentropy,
              metrics=['accuracy'])

학습시키기 위해서는 Optimer , Loss, Metrics 가 필요하다.

opt : 학습을 어떤 방식으로 시킬 것인지
loss : 모델이 학습해나가야 할 방향 (이 경우는 확률분포)
metrics : [accuracy, precision, recall]

아직은 실행하기 전이다. 지금은 사전 작업만 거친 상태

BATCH_SIZE = 32
SHUFFLE_BUFFER_SIZE = 1000

train_batches = train.shuffle(SHUFFLE_BUFFER_SIZE).batch(BATCH_SIZE)
validation_batches = validation.batch(BATCH_SIZE)
test_batches = test.batch(BATCH_SIZE)

이렇게 train 데이터를 통으로 넣지 않고 32개씩 끊어 넣는 이유는 랜덤한 32개의 사진들을 묶어 여러개의 Decision Tree 를 만들어 ansamble 하기 위함

train_batches 의 데이터를 확인해보자

for image_batch, label_batch in train_batches.take(1):
    pass

image_batch.shape, label_batch.shape

(TensorShape([32, 160, 160, 3]), TensorShape([32]))

모델 학습 전에 초기 모델의 성능을 테스트해볼까? validation data 로 모델을 평가해보자
20번의 예측을 해보고 loss 와 accuracy를 구해보자

validation_steps = 20
loss0, accuracy0 = model.evaluate(validation_batches, steps=validation_steps)

print("initial loss: {:.2f}".format(loss0))
print("initial accuracy: {:.2f}".format(accuracy0))

10/20 [==============>...............] - ETA: 0s - loss: 0.6924 - accuracy: 0.5156

Corrupt JPEG data: 162 extraneous bytes before marker 0xd9


20/20 [==============================] - 3s 32ms/step - loss: 0.6919 - accuracy: 0.5172
initial loss: 0.69
initial accuracy: 0.52


Corrupt JPEG data: 252 extraneous bytes before marker 0xd9

보면 아무것도 하기 전에 모델 evaluate 는 걍 뭐 아무것도 모른다. 이제 이걸 학습시켜보자

EPOCHS = 10
history = model.fit(train_batches,
                    epochs=EPOCHS,
                    validation_data=validation_batches)

Epoch 1/10
170/582 [=======>......................] - ETA: 18s - loss: 0.5444 - accuracy: 0.7250

Corrupt JPEG data: 99 extraneous bytes before marker 0xd9


202/582 [=========>....................] - ETA: 17s - loss: 0.5408 - accuracy: 0.7274

Warning: unknown JFIF revision number 0.00


212/582 [=========>....................] - ETA: 16s - loss: 0.5427 - accuracy: 0.7258

Corrupt JPEG data: 396 extraneous bytes before marker 0xd9


284/582 [=============>................] - ETA: 13s - loss: 0.5408 - accuracy: 0.7272

Corrupt JPEG data: 65 extraneous bytes before marker 0xd9


492/582 [========================>.....] - ETA: 3s - loss: 0.5280 - accuracy: 0.7363

Corrupt JPEG data: 2226 extraneous bytes before marker 0xd9


504/582 [========================>.....] - ETA: 3s - loss: 0.5277 - accuracy: 0.7360

Corrupt JPEG data: 128 extraneous bytes before marker 0xd9


516/582 [=========================>....] - ETA: 2s - loss: 0.5271 - accuracy: 0.7372

Corrupt JPEG data: 239 extraneous bytes before marker 0xd9


542/582 [==========================>...] - ETA: 1s - loss: 0.5269 - accuracy: 0.7374

Corrupt JPEG data: 1153 extraneous bytes before marker 0xd9


550/582 [===========================>..] - ETA: 1s - loss: 0.5271 - accuracy: 0.7372

Corrupt JPEG data: 228 extraneous bytes before marker 0xd9


582/582 [==============================] - ETA: 0s - loss: 0.5259 - accuracy: 0.7389

Corrupt JPEG data: 162 extraneous bytes before marker 0xd9
Corrupt JPEG data: 252 extraneous bytes before marker 0xd9
Corrupt JPEG data: 214 extraneous bytes before marker 0xd9


582/582 [==============================] - 29s 48ms/step - loss: 0.5259 - accuracy: 0.7389 - val_loss: 0.5102 - val_accuracy: 0.7502
Epoch 2/10
169/582 [=======>......................] - ETA: 18s - loss: 0.4561 - accuracy: 0.7898

Corrupt JPEG data: 99 extraneous bytes before marker 0xd9


201/582 [=========>....................] - ETA: 16s - loss: 0.4553 - accuracy: 0.7886

Warning: unknown JFIF revision number 0.00


213/582 [=========>....................] - ETA: 16s - loss: 0.4548 - accuracy: 0.7893

Corrupt JPEG data: 396 extraneous bytes before marker 0xd9


283/582 [=============>................] - ETA: 13s - loss: 0.4483 - accuracy: 0.7911

Corrupt JPEG data: 65 extraneous bytes before marker 0xd9


492/582 [========================>.....] - ETA: 3s - loss: 0.4446 - accuracy: 0.7943

Corrupt JPEG data: 2226 extraneous bytes before marker 0xd9


506/582 [=========================>....] - ETA: 3s - loss: 0.4450 - accuracy: 0.7947

Corrupt JPEG data: 128 extraneous bytes before marker 0xd9


516/582 [=========================>....] - ETA: 2s - loss: 0.4445 - accuracy: 0.7951

Corrupt JPEG data: 239 extraneous bytes before marker 0xd9


542/582 [==========================>...] - ETA: 1s - loss: 0.4437 - accuracy: 0.7953

Corrupt JPEG data: 1153 extraneous bytes before marker 0xd9


548/582 [===========================>..] - ETA: 1s - loss: 0.4433 - accuracy: 0.7952

Corrupt JPEG data: 228 extraneous bytes before marker 0xd9


580/582 [============================>.] - ETA: 0s - loss: 0.4425 - accuracy: 0.7956

Corrupt JPEG data: 162 extraneous bytes before marker 0xd9
Corrupt JPEG data: 252 extraneous bytes before marker 0xd9
Corrupt JPEG data: 214 extraneous bytes before marker 0xd9


582/582 [==============================] - 28s 47ms/step - loss: 0.4422 - accuracy: 0.7958 - val_loss: 0.5503 - val_accuracy: 0.7386
Epoch 3/10
169/582 [=======>......................] - ETA: 18s - loss: 0.3937 - accuracy: 0.8229

Corrupt JPEG data: 99 extraneous bytes before marker 0xd9


201/582 [=========>....................] - ETA: 16s - loss: 0.3927 - accuracy: 0.8232

Warning: unknown JFIF revision number 0.00


213/582 [=========>....................] - ETA: 16s - loss: 0.3927 - accuracy: 0.8234

Corrupt JPEG data: 396 extraneous bytes before marker 0xd9


283/582 [=============>................] - ETA: 13s - loss: 0.3870 - accuracy: 0.8266

Corrupt JPEG data: 65 extraneous bytes before marker 0xd9


492/582 [========================>.....] - ETA: 3s - loss: 0.3836 - accuracy: 0.8281

Corrupt JPEG data: 2226 extraneous bytes before marker 0xd9


506/582 [=========================>....] - ETA: 3s - loss: 0.3843 - accuracy: 0.8278

Corrupt JPEG data: 128 extraneous bytes before marker 0xd9


516/582 [=========================>....] - ETA: 2s - loss: 0.3836 - accuracy: 0.8281

Corrupt JPEG data: 239 extraneous bytes before marker 0xd9


543/582 [==========================>...] - ETA: 1s - loss: 0.3820 - accuracy: 0.8288

Corrupt JPEG data: 1153 extraneous bytes before marker 0xd9


549/582 [===========================>..] - ETA: 1s - loss: 0.3814 - accuracy: 0.8292

Corrupt JPEG data: 228 extraneous bytes before marker 0xd9


581/582 [============================>.] - ETA: 0s - loss: 0.3800 - accuracy: 0.8289

Corrupt JPEG data: 162 extraneous bytes before marker 0xd9
Corrupt JPEG data: 252 extraneous bytes before marker 0xd9
Corrupt JPEG data: 214 extraneous bytes before marker 0xd9


582/582 [==============================] - 28s 47ms/step - loss: 0.3799 - accuracy: 0.8290 - val_loss: 0.4954 - val_accuracy: 0.7674
Epoch 4/10
169/582 [=======>......................] - ETA: 18s - loss: 0.3449 - accuracy: 0.8460

Corrupt JPEG data: 99 extraneous bytes before marker 0xd9


201/582 [=========>....................] - ETA: 16s - loss: 0.3404 - accuracy: 0.8469

Warning: unknown JFIF revision number 0.00


213/582 [=========>....................] - ETA: 16s - loss: 0.3395 - accuracy: 0.8480

Corrupt JPEG data: 396 extraneous bytes before marker 0xd9


283/582 [=============>................] - ETA: 12s - loss: 0.3357 - accuracy: 0.8520

Corrupt JPEG data: 65 extraneous bytes before marker 0xd9


492/582 [========================>.....] - ETA: 3s - loss: 0.3270 - accuracy: 0.8581

Corrupt JPEG data: 2226 extraneous bytes before marker 0xd9


506/582 [=========================>....] - ETA: 3s - loss: 0.3273 - accuracy: 0.8581

Corrupt JPEG data: 128 extraneous bytes before marker 0xd9


516/582 [=========================>....] - ETA: 2s - loss: 0.3260 - accuracy: 0.8586

Corrupt JPEG data: 239 extraneous bytes before marker 0xd9


542/582 [==========================>...] - ETA: 1s - loss: 0.3256 - accuracy: 0.8590

Corrupt JPEG data: 1153 extraneous bytes before marker 0xd9


550/582 [===========================>..] - ETA: 1s - loss: 0.3252 - accuracy: 0.8593

Corrupt JPEG data: 228 extraneous bytes before marker 0xd9


580/582 [============================>.] - ETA: 0s - loss: 0.3244 - accuracy: 0.8602

Corrupt JPEG data: 162 extraneous bytes before marker 0xd9
Corrupt JPEG data: 252 extraneous bytes before marker 0xd9
Corrupt JPEG data: 214 extraneous bytes before marker 0xd9


582/582 [==============================] - 28s 47ms/step - loss: 0.3244 - accuracy: 0.8602 - val_loss: 0.4858 - val_accuracy: 0.7825
Epoch 5/10
170/582 [=======>......................] - ETA: 18s - loss: 0.2913 - accuracy: 0.8765

Corrupt JPEG data: 99 extraneous bytes before marker 0xd9


202/582 [=========>....................] - ETA: 16s - loss: 0.2866 - accuracy: 0.8790

Warning: unknown JFIF revision number 0.00


212/582 [=========>....................] - ETA: 16s - loss: 0.2868 - accuracy: 0.8791

Corrupt JPEG data: 396 extraneous bytes before marker 0xd9


284/582 [=============>................] - ETA: 12s - loss: 0.2804 - accuracy: 0.8825

Corrupt JPEG data: 65 extraneous bytes before marker 0xd9


493/582 [========================>.....] - ETA: 3s - loss: 0.2756 - accuracy: 0.8869

Corrupt JPEG data: 2226 extraneous bytes before marker 0xd9


505/582 [=========================>....] - ETA: 3s - loss: 0.2740 - accuracy: 0.8876

Corrupt JPEG data: 128 extraneous bytes before marker 0xd9


515/582 [=========================>....] - ETA: 2s - loss: 0.2733 - accuracy: 0.8877

Corrupt JPEG data: 239 extraneous bytes before marker 0xd9


543/582 [==========================>...] - ETA: 1s - loss: 0.2732 - accuracy: 0.8886

Corrupt JPEG data: 1153 extraneous bytes before marker 0xd9


549/582 [===========================>..] - ETA: 1s - loss: 0.2727 - accuracy: 0.8889

Corrupt JPEG data: 228 extraneous bytes before marker 0xd9


581/582 [============================>.] - ETA: 0s - loss: 0.2720 - accuracy: 0.8886

Corrupt JPEG data: 162 extraneous bytes before marker 0xd9
Corrupt JPEG data: 252 extraneous bytes before marker 0xd9
Corrupt JPEG data: 214 extraneous bytes before marker 0xd9


582/582 [==============================] - 28s 46ms/step - loss: 0.2721 - accuracy: 0.8885 - val_loss: 0.4933 - val_accuracy: 0.7919
Epoch 6/10
169/582 [=======>......................] - ETA: 18s - loss: 0.2373 - accuracy: 0.9051

Corrupt JPEG data: 99 extraneous bytes before marker 0xd9


201/582 [=========>....................] - ETA: 16s - loss: 0.2335 - accuracy: 0.9083

Warning: unknown JFIF revision number 0.00


211/582 [=========>....................] - ETA: 16s - loss: 0.2331 - accuracy: 0.9083

Corrupt JPEG data: 396 extraneous bytes before marker 0xd9


283/582 [=============>................] - ETA: 13s - loss: 0.2268 - accuracy: 0.9110

Corrupt JPEG data: 65 extraneous bytes before marker 0xd9


493/582 [========================>.....] - ETA: 3s - loss: 0.2216 - accuracy: 0.9123

Corrupt JPEG data: 2226 extraneous bytes before marker 0xd9


505/582 [=========================>....] - ETA: 3s - loss: 0.2209 - accuracy: 0.9127

Corrupt JPEG data: 128 extraneous bytes before marker 0xd9


516/582 [=========================>....] - ETA: 2s - loss: 0.2196 - accuracy: 0.9133

Corrupt JPEG data: 239 extraneous bytes before marker 0xd9


542/582 [==========================>...] - ETA: 1s - loss: 0.2178 - accuracy: 0.9147

Corrupt JPEG data: 1153 extraneous bytes before marker 0xd9


548/582 [===========================>..] - ETA: 1s - loss: 0.2171 - accuracy: 0.9150

Corrupt JPEG data: 228 extraneous bytes before marker 0xd9


580/582 [============================>.] - ETA: 0s - loss: 0.2163 - accuracy: 0.9152

Corrupt JPEG data: 162 extraneous bytes before marker 0xd9
Corrupt JPEG data: 252 extraneous bytes before marker 0xd9
Corrupt JPEG data: 214 extraneous bytes before marker 0xd9


582/582 [==============================] - 29s 47ms/step - loss: 0.2162 - accuracy: 0.9153 - val_loss: 0.6168 - val_accuracy: 0.7601
Epoch 7/10
169/582 [=======>......................] - ETA: 18s - loss: 0.1858 - accuracy: 0.9310

Corrupt JPEG data: 99 extraneous bytes before marker 0xd9


201/582 [=========>....................] - ETA: 16s - loss: 0.1836 - accuracy: 0.9331

Warning: unknown JFIF revision number 0.00


213/582 [=========>....................] - ETA: 16s - loss: 0.1820 - accuracy: 0.9341

Corrupt JPEG data: 396 extraneous bytes before marker 0xd9


283/582 [=============>................] - ETA: 13s - loss: 0.1783 - accuracy: 0.9361

Corrupt JPEG data: 65 extraneous bytes before marker 0xd9


493/582 [========================>.....] - ETA: 3s - loss: 0.1724 - accuracy: 0.9365

Corrupt JPEG data: 2226 extraneous bytes before marker 0xd9


505/582 [=========================>....] - ETA: 3s - loss: 0.1717 - accuracy: 0.9369

Corrupt JPEG data: 128 extraneous bytes before marker 0xd9


515/582 [=========================>....] - ETA: 2s - loss: 0.1709 - accuracy: 0.9374

Corrupt JPEG data: 239 extraneous bytes before marker 0xd9


543/582 [==========================>...] - ETA: 1s - loss: 0.1699 - accuracy: 0.9378

Corrupt JPEG data: 1153 extraneous bytes before marker 0xd9


549/582 [===========================>..] - ETA: 1s - loss: 0.1698 - accuracy: 0.9378

Corrupt JPEG data: 228 extraneous bytes before marker 0xd9


581/582 [============================>.] - ETA: 0s - loss: 0.1678 - accuracy: 0.9382

Corrupt JPEG data: 162 extraneous bytes before marker 0xd9
Corrupt JPEG data: 252 extraneous bytes before marker 0xd9
Corrupt JPEG data: 214 extraneous bytes before marker 0xd9


582/582 [==============================] - 28s 47ms/step - loss: 0.1677 - accuracy: 0.9382 - val_loss: 0.6220 - val_accuracy: 0.7627
Epoch 8/10
170/582 [=======>......................] - ETA: 18s - loss: 0.1322 - accuracy: 0.9537

Corrupt JPEG data: 99 extraneous bytes before marker 0xd9


202/582 [=========>....................] - ETA: 16s - loss: 0.1317 - accuracy: 0.9547

Warning: unknown JFIF revision number 0.00


212/582 [=========>....................] - ETA: 16s - loss: 0.1293 - accuracy: 0.9558

Corrupt JPEG data: 396 extraneous bytes before marker 0xd9


284/582 [=============>................] - ETA: 13s - loss: 0.1267 - accuracy: 0.9558

Corrupt JPEG data: 65 extraneous bytes before marker 0xd9


493/582 [========================>.....] - ETA: 3s - loss: 0.1241 - accuracy: 0.9564

Corrupt JPEG data: 2226 extraneous bytes before marker 0xd9


505/582 [=========================>....] - ETA: 3s - loss: 0.1237 - accuracy: 0.9566

Corrupt JPEG data: 128 extraneous bytes before marker 0xd9


515/582 [=========================>....] - ETA: 2s - loss: 0.1235 - accuracy: 0.9567

Corrupt JPEG data: 239 extraneous bytes before marker 0xd9


542/582 [==========================>...] - ETA: 1s - loss: 0.1223 - accuracy: 0.9574

Corrupt JPEG data: 1153 extraneous bytes before marker 0xd9


550/582 [===========================>..] - ETA: 1s - loss: 0.1219 - accuracy: 0.9577

Corrupt JPEG data: 228 extraneous bytes before marker 0xd9


580/582 [============================>.] - ETA: 0s - loss: 0.1210 - accuracy: 0.9580

Corrupt JPEG data: 162 extraneous bytes before marker 0xd9
Corrupt JPEG data: 252 extraneous bytes before marker 0xd9
Corrupt JPEG data: 214 extraneous bytes before marker 0xd9


582/582 [==============================] - 28s 47ms/step - loss: 0.1209 - accuracy: 0.9581 - val_loss: 0.6818 - val_accuracy: 0.7623
Epoch 9/10
170/582 [=======>......................] - ETA: 18s - loss: 0.0973 - accuracy: 0.9688

Corrupt JPEG data: 99 extraneous bytes before marker 0xd9


202/582 [=========>....................] - ETA: 16s - loss: 0.0932 - accuracy: 0.9711

Warning: unknown JFIF revision number 0.00


212/582 [=========>....................] - ETA: 16s - loss: 0.0942 - accuracy: 0.9704

Corrupt JPEG data: 396 extraneous bytes before marker 0xd9


284/582 [=============>................] - ETA: 13s - loss: 0.0915 - accuracy: 0.9712

Corrupt JPEG data: 65 extraneous bytes before marker 0xd9


492/582 [========================>.....] - ETA: 3s - loss: 0.0893 - accuracy: 0.9715

Corrupt JPEG data: 2226 extraneous bytes before marker 0xd9


506/582 [=========================>....] - ETA: 3s - loss: 0.0888 - accuracy: 0.9719

Corrupt JPEG data: 128 extraneous bytes before marker 0xd9


516/582 [=========================>....] - ETA: 2s - loss: 0.0883 - accuracy: 0.9721

Corrupt JPEG data: 239 extraneous bytes before marker 0xd9


542/582 [==========================>...] - ETA: 1s - loss: 0.0880 - accuracy: 0.9722

Corrupt JPEG data: 1153 extraneous bytes before marker 0xd9


550/582 [===========================>..] - ETA: 1s - loss: 0.0874 - accuracy: 0.9725

Corrupt JPEG data: 228 extraneous bytes before marker 0xd9


580/582 [============================>.] - ETA: 0s - loss: 0.0875 - accuracy: 0.9725

Corrupt JPEG data: 162 extraneous bytes before marker 0xd9
Corrupt JPEG data: 252 extraneous bytes before marker 0xd9
Corrupt JPEG data: 214 extraneous bytes before marker 0xd9


582/582 [==============================] - 28s 47ms/step - loss: 0.0874 - accuracy: 0.9725 - val_loss: 0.6500 - val_accuracy: 0.7863
Epoch 10/10
170/582 [=======>......................] - ETA: 18s - loss: 0.0692 - accuracy: 0.9783

Corrupt JPEG data: 99 extraneous bytes before marker 0xd9


202/582 [=========>....................] - ETA: 17s - loss: 0.0681 - accuracy: 0.9785

Warning: unknown JFIF revision number 0.00


212/582 [=========>....................] - ETA: 16s - loss: 0.0674 - accuracy: 0.9788

Corrupt JPEG data: 396 extraneous bytes before marker 0xd9


284/582 [=============>................] - ETA: 13s - loss: 0.0648 - accuracy: 0.9798

Corrupt JPEG data: 65 extraneous bytes before marker 0xd9


493/582 [========================>.....] - ETA: 3s - loss: 0.0626 - accuracy: 0.9816

Corrupt JPEG data: 2226 extraneous bytes before marker 0xd9


505/582 [=========================>....] - ETA: 3s - loss: 0.0624 - accuracy: 0.9817

Corrupt JPEG data: 128 extraneous bytes before marker 0xd9


515/582 [=========================>....] - ETA: 2s - loss: 0.0621 - accuracy: 0.9820

Corrupt JPEG data: 239 extraneous bytes before marker 0xd9


543/582 [==========================>...] - ETA: 1s - loss: 0.0610 - accuracy: 0.9826

Corrupt JPEG data: 1153 extraneous bytes before marker 0xd9


549/582 [===========================>..] - ETA: 1s - loss: 0.0615 - accuracy: 0.9825

Corrupt JPEG data: 228 extraneous bytes before marker 0xd9


581/582 [============================>.] - ETA: 0s - loss: 0.0614 - accuracy: 0.9827

Corrupt JPEG data: 162 extraneous bytes before marker 0xd9
Corrupt JPEG data: 252 extraneous bytes before marker 0xd9
Corrupt JPEG data: 214 extraneous bytes before marker 0xd9


582/582 [==============================] - 29s 47ms/step - loss: 0.0613 - accuracy: 0.9827 - val_loss: 0.6904 - val_accuracy: 0.7889

다음은 학습시킨 모델에 대한 그래프 보고이다

for image_batch, label_batch in test_batches.take(1):
    images = image_batch
    labels = label_batch
    predictions = model.predict(image_batch)
    pass

predictions

array([[9.9999702e-01, 2.9417115e-06],
       [8.3842850e-01, 1.6157144e-01],
       [7.4218982e-01, 2.5781012e-01],
       [9.9970847e-01, 2.9155653e-04],
       [9.9640131e-01, 3.5986665e-03],
       [7.8427315e-02, 9.2157269e-01],
       [1.8649189e-02, 9.8135078e-01],
       [9.1933328e-01, 8.0666728e-02],
       [9.4125561e-02, 9.0587437e-01],
       [7.8091651e-02, 9.2190832e-01],
       [1.1057696e-01, 8.8942307e-01],
       [9.2630666e-01, 7.3693395e-02],
       [9.9996006e-01, 3.9985713e-05],
       [4.6824862e-05, 9.9995315e-01],
       [9.9207205e-01, 7.9279514e-03],
       [9.9890709e-01, 1.0929310e-03],
       [3.1051392e-02, 9.6894866e-01],
       [5.8752208e-09, 1.0000000e+00],
       [8.3172768e-01, 1.6827232e-01],
       [9.9952376e-01, 4.7630019e-04],
       [7.9531687e-01, 2.0468311e-01],
       [9.9072027e-01, 9.2797335e-03],
       [9.9999344e-01, 6.5057743e-06],
       [9.1565454e-01, 8.4345400e-02],
       [9.9570221e-01, 4.2977203e-03],
       [1.0600092e-02, 9.8939985e-01],
       [9.9997663e-01, 2.3377719e-05],
       [1.0107538e-01, 8.9892459e-01],
       [9.9934202e-01, 6.5792818e-04],
       [9.9927968e-01, 7.2028313e-04],
       [9.9999619e-01, 3.7975171e-06],
       [2.8957015e-01, 7.1042985e-01]], dtype=float32)

이것이 바로 우리의 정확도이다.

import numpy as np

predictions = np.argmax(predictions, axis=1)
predictions

array([0, 0, 0, 0, 0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 0, 0, 0,
       0, 0, 0, 1, 0, 1, 0, 0, 0, 1])

이제 32장의 image 와 32개의 label , 32개의 prediction 을 얻었다 최종 확인을 해보자

plt.figure(figsize=(20, 12))

for idx, (image, label, prediction) in enumerate(zip(images, labels, predictions)):
    plt.subplot(4, 8, idx+1)
    image = (image + 1) / 2
    plt.imshow(image)
    correct = label == prediction
    title = f'real: {label} / pred :{prediction}\n {correct}!'
    if not correct:
        plt.title(title, fontdict={'color': 'red'})
    else:
        plt.title(title, fontdict={'color': 'blue'})
    plt.axis('off')

png

count = 0   # 정답을 맞춘 개수
for image, label, prediction in zip(images, labels, predictions):
    correct = label == prediction
    if correct:
        count = count + 1

print(count / 32 * 100)

71.875

중간에 들어가있는 딥러닝 모델이 단순한 구조이기 때문에
좋은 결과가 나오고 있지는 않다.

다만 중간의 모델구조를 변경한다면 더 좋은 결과를
끌어낼 수 있을 것이다.

강아지 & 고양이 분류 프로그램 제작 !
- Content
1. 데이터 전처리
2 모델 생성 및 학습