더북(TheBook)

앞서 생성한 클래스(AlexNet)를 호출하여 AlexNet이라는 모델을 생성합니다. 이때 AlexNet 클래스에 전달되는 입력 값은 (100,100,3)의 형태이고, 출력은 개와 고양이를 표현하는 값 2가 됩니다.

코드 6-10 모델 생성

model = AlexNet((100,100,3), num_classes)
model.summary()

다음은 AlexNet에 대한 네트워크를 출력한 결과입니다.

Model: "alex_net"
________________________________________________________________
Layer (type)                 Output Shape              Param #
================================================================
conv2d_2 (Conv2D)            (None, 23, 23, 96)        34944
________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 11, 11, 96)        0
________________________________________________________________
conv2d_3 (Conv2D)            (None, 11, 11, 256)       614656
________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 5, 5, 256)         0
________________________________________________________________
conv2d_4 (Conv2D)            (None, 5, 5, 384)         885120
________________________________________________________________
conv2d_5 (Conv2D)            (None, 5, 5, 384)         1327488
________________________________________________________________
conv2d_6 (Conv2D)            (None, 5, 5, 256)         884992
________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 2, 2, 256)         0
________________________________________________________________
flatten_1 (Flatten)          (None, 1024)              0
________________________________________________________________
dense_3 (Dense)              (None, 4096)              4198400
________________________________________________________________
dense_4 (Dense)              (None, 4096)              16781312
________________________________________________________________
dense_5 (Dense)              (None, 1000)              4097000
________________________________________________________________
dense_6 (Dense)              (None, 2)                 2002
================================================================
Total params: 28,825,914
Trainable params: 28,825,914
Non-trainable params: 0
________________________________________________________________
신간 소식 구독하기
뉴스레터에 가입하시고 이메일로 신간 소식을 받아 보세요.