이론과 관련된 내용은 여기를 참고하시면 됩니다.

Conv 2D 층

from tensorflow.keras.layers import Input, Conv2D, MaxPooling2D
from tensorflow.keras.models import Model

input_tensor = Input(shape=(28, 28, 1))
x = Conv2D(filters=4, kernel_size=3, strides=1, padding='same', activation='relu')(input_tensor)
print('x type:', type(x), 'x:', x)
----------------------------------------------------------------------------------------------
x type: <class 'tensorflow.python.keras.engine.keras_tensor.KerasTensor'> x: KerasTensor(type_spec=TensorSpec(shape=(None, 28, 28, 4), dtype=tf.float32, name=None), name='conv2d/Relu:0', description="created by layer 'conv2d'")

Pooling 층

input_tensor = Input(shape=(28, 28, 1))
x = Conv2D(filters=16, kernel_size=3, strides=1, padding='same', activation='relu')(input_tensor)
x = MaxPooling2D(2)(x)
print(x)
---------------------------------------------------------------------------------------------------
KerasTensor(type_spec=TensorSpec(shape=(None, 14, 14, 16), dtype=tf.float32, name=None), name='max_pooling2d/MaxPool:0', description="created by layer 'max_pooling2d'")

CNN 모델 생성하기

input_tensor = Input(shape=(28, 28, 1))
x = Conv2D(filters=32, kernel_size=3, strides=1, padding='same', activation='relu')(input_tensor)
x = Conv2D(filters=64, kernel_size=3, activation='relu')(x)
x = MaxPooling2D(2)(x)

model = Model(inputs=input_tensor, outputs=x)
model.summary()
-----------------------------------------------------------------------------------------------------
Model: "model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_4 (InputLayer)         [(None, 28, 28, 1)]       0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 28, 28, 32)        320       
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 26, 26, 64)        18496     
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 13, 13, 64)        0         
=================================================================
Total params: 18,816
Trainable params: 18,816
Non-trainable params: 0
_________________________________________________________________

from tensorflow.keras.layers import Dense, Flatten

input_tensor = Input(shape=(28, 28, 1))
x = Conv2D(filters=32, kernel_size=3, strides=1, padding='same', activation='relu')(input_tensor)
x = Conv2D(filters=64, kernel_size=3, activation='relu')(x)
x = MaxPooling2D(2)(x)

# 3차원으로 되어있는 Feature map 결과를 Fully Connected 연결하기 위해서는 Flatten()을 적용해야함. 
x = Flatten()(x)
x = Dense(100, activation='relu')(x)
output = Dense(10, activation='softmax')(x)
model = Model(inputs=input_tensor, outputs=output)
model.summary()
-----------------------------------------------------------------------------------------------
Model: "model_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_5 (InputLayer)         [(None, 28, 28, 1)]       0         
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 28, 28, 32)        320       
_________________________________________________________________
conv2d_6 (Conv2D)            (None, 26, 26, 64)        18496     
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 13, 13, 64)        0         
_________________________________________________________________
flatten (Flatten)            (None, 10816)             0         
_________________________________________________________________
dense (Dense)                (None, 100)               1081700   
_________________________________________________________________
dense_1 (Dense)              (None, 10)                1010      
=================================================================
Total params: 1,101,526
Trainable params: 1,101,526
Non-trainable params: 0

Stride가 1이고 Padding이 없는 경우

• I는 입력 Feature Map의 크기, F는 Filter의 크기(Kernel size), P는 Padding(정수), S는 Strides(정수)
• O = (I - F + 2P)/2 + 1 = (5 - 3 + 0 )/1 + 1 = 3

from tensorflow.keras.layers import Input, Conv2D, MaxPooling2D
from tensorflow.keras.models import Model

input_tensor = Input(shape=(5, 5, 1))
x = Conv2D(filters=1, kernel_size=3, strides=1)(input_tensor)
print('x.shape:', x.shape)
--------------------------------------------------------------
x.shape: (None, 3, 3, 1)