[시각 지능] CNN Basic Architecture

CNN Basic Architecture

 

특징 추출이라고 하는 프로세스에서 분석을 위해 이미지의 다양한 특징을 분리하고 식별하는 컨볼루션 도구이다.  특징 추출 네트워크는 여러 쌍의 컨볼루션 또는 풀링 레이어로 구성된다.  컨볼루션 프로세스의 출력을 활용하고 이전 단계에서 추출된 특징을 기반으로 이미지의 클래스를 예측하는 완전 연결 계층이다. 특징 추출의 이 CNN 모델은 데이터 세트에 존재하는 특징의 수를 줄이는 것을 목표로 한다. 원래 기능 세트에 포함된 기존 기능을 요약하는 새 기능을 생성한다. CNN 아키텍처 다이어그램 에 표시된 것처럼 많은 CNN 레이어가 있다.

import tensorflow as tf

from tensorflow.keras.layers import Flatten, Dense, Conv2D, MaxPool2D
from tensorflow.keras.models import Sequential
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.datasets import mnist

import numpy as np
from datetime import datetime
import matplotlib.pyplot as plt

print(tf.__version__)

2.8.2

(x_train, t_train), (x_test, t_test) = mnist.load_data()

x_train = x_train / 255.0
x_test = x_test / 255.0

print('x_train.shape = ', x_train.shape, ' , x_test.shape = ', x_test.shape)
print('t_train.shape = ', t_train.shape, ' , t_test.shape = ', t_test.shape)

Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz
11493376/11490434 [==============================] - 0s 0us/step
11501568/11490434 [==============================] - 0s 0us/step
x_train.shape =  (60000, 28, 28)  , x_test.shape =  (10000, 28, 28)
t_train.shape =  (60000,)  , t_test.shape =  (10000,)

# 데이터 출력
plt.figure(figsize=(10,10))

for index in range(25):
    plt.subplot(5, 5, index+1)
    plt.xticks([])
    plt.yticks([])
    plt.grid(False)
    plt.imshow(x_train[index], cmap='gray')
    plt.xlabel(str(t_train[index]))

plt.show()

 

one-hot encoding

 

# one hot encoding
t_train = to_categorical(t_train, 10)
t_test = to_categorical(t_test, 10)

model = Sequential()

model.add(Conv2D(input_shape=(28,28,1), 
                 kernel_size=3, filters=32, 
                 strides=(1,1), activation='relu', padding='SAME'))
model.add(MaxPool2D(pool_size=(2,2), padding='SAME'))
model.add(Flatten())
model.add(Dense(10, activation='softmax'))

# model compile
# one hot encoding 방식이기 때문에, loss=categorical_crossentropy 정의
model.compile(optimizer=Adam(learning_rate=0.001), 
              loss='categorical_crossentropy', metrics=['accuracy'])

model.summary()

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 28, 28, 32)        320       
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 14, 14, 32)        0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 6272)              0         
_________________________________________________________________
dense_1 (Dense)              (None, 10)                62730     
=================================================================
Total params: 63,050
Trainable params: 63,050
Non-trainable params: 0
_________________________________________________________________

 

sparse

 

# model compile
# one hot encoding 방식이 아니기 때문에, loss=sparse_categorical_crossentropy 정의
model = Sequential()

model.add(Conv2D(input_shape = (28, 28, 1),
          kernel_size = 3, filters = 32,
          strides = (1, 1), activation = 'relu', use_bias = True, padding = 'SAME'))
model.add(MaxPool2D(pool_size = (2, 2), padding = 'SAME'))
model.add(Flatten())
model.add(Dense(10, activation = 'softmax'))

model.compile(optimizer = Adam(learning_rate = 0.001), loss = 'sparse_categorical_crossentropy', metrics = ['accuracy'])

model.summary()

Model: "sequential_7"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d_4 (Conv2D)           (None, 28, 28, 32)        320       
                                                                 
 max_pooling2d_3 (MaxPooling  (None, 14, 14, 32)       0         
 2D)                                                             
                                                                 
 flatten_2 (Flatten)         (None, 6272)              0         
                                                                 
 dense_2 (Dense)             (None, 10)                62730     
                                                                 
=================================================================
Total params: 63,050
Trainable params: 63,050
Non-trainable params: 0
_________________________________________________________________

start_time = datetime.now()

hist = model.fit(x_train.reshape(-1,28,28,1), t_train, 
                 batch_size=50, epochs=50, validation_split=0.2)

end_time = datetime.now()

print('\n\nElapsed Time => ', end_time - start_time)

model.evaluate(x_test.reshape(-1, 28, 28, 1), t_test)

313/313 [==============================] - 3s 8ms/step - loss: 0.1100 - accuracy: 0.9825
[0.10999356955289841, 0.9825000166893005]

plt.title('Loss')
plt.grid()
plt.xlabel('epochs')
plt.ylabel('loss')
plt.plot(hist.history['loss'], label='train loss')
plt.plot(hist.history['val_loss'], label='validation loss')
plt.legend(loc='best')

plt.show()

plt.title('Accuracy')
plt.grid()
plt.xlabel('epochs')
plt.ylabel('accuracy')
plt.plot(hist.history['accuracy'], label='train accuracy')
plt.plot(hist.history['val_accuracy'], label='validation accuracy')
plt.legend(loc='best')

plt.show()

 

테스트 데이터의 정답과 예측 값 비교

 

rand_idx = np.random.randint(len(x_test))

print(rand_idx)

predicted_val = model.predict(x_test[rand_idx].reshape(-1,28,28,1))

print(type(predicted_val), predicted_val.shape)

print(predicted_val)

print(np.argmax(predicted_val), t_test[rand_idx])

3168
<class 'numpy.ndarray'> (1, 10)
[[2.9477740e-31 1.0732472e-25 1.2657751e-20 3.5341849e-13 1.3210143e-16
  1.9856330e-19 1.2783073e-30 2.5408397e-16 5.2557976e-16 1.0000000e+00]]
9 9

# x_test를 4차원 tensor로 변환
x_test = x_test.reshape(-1, 28, 28, 1)

print(x_test.shape)

(10000, 28, 28, 1)

# 중복하지 않고 random index 추출
rand_idx = np.random.choice(len(x_test), 3, replace=False)

print(rand_idx)

rand_idx = np.random.choice(len(x_test), 3, replace=False)

print(rand_idx)

# axis=-1 이용하여 행 기준으로 predict 결과 추출
predicted_val = model.predict(x_test[rand_idx])

print(type(predicted_val), predicted_val.shape)

print(np.argmax(predicted_val, axis=-1), t_test[rand_idx])

# np.array() 이용하여 4차원 tensor 변환
test_data_list = [ x_test[idx]  for idx in rand_idx  ]

print(len(test_data_list))

test_data_array = np.array(test_data_list)

print(test_data_array.shape)

3
(3, 28, 28, 1)

predicted_val = model.predict(test_data_array)

print(type(predicted_val), predicted_val.shape)

print(np.argmax(predicted_val, axis=-1), t_test[rand_idx])