理解model.summary Keras

Question

我正在努力理解Keras中的model.summary()。我有以下卷积神经网络。第一次卷积的值是：

conv2d_4 (Conv2D)            (None, 148, 148, 16)      448 

148和448来自哪里？

码

image_input = layers.Input(shape=(150, 150, 3))
x = layers.Conv2D(16, 3, activation='relu')(image_input)

x = layers.MaxPooling2D(2)(x)
x = layers.Conv2D(32, 3, activation='relu')(x)

x = layers.MaxPooling2D(2)(x)
x = layers.Conv2D(64, 3, activation='relu')(x)

x = layers.MaxPooling2D(2)(x)
x = layers.Flatten()(x)
x = layers.Dense(512, activation='relu')(x)
output = layers.Dense(1, activation='sigmoid')(x)

# Keras Model definition
# input = input feature map
# output = input feature map + stacked convolution/maxpooling layers + fully connected layer + sigmoid output layer
model = Model(image_input, output)
model.summary()

输出

Layer (type)                 Output Shape              Param #   
=================================================================
input_2 (InputLayer)         (None, 150, 150, 3)       0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 148, 148, 16)      448       
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 74, 74, 16)        0         
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 72, 72, 32)        4640      
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 36, 36, 32)        0         
_________________________________________________________________
conv2d_6 (Conv2D)            (None, 34, 34, 64)        18496     
_________________________________________________________________
max_pooling2d_6 (MaxPooling2 (None, 17, 17, 64)        0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 18496)             0         
_________________________________________________________________
dense_1 (Dense)              (None, 512)               9470464   
_________________________________________________________________
dense_2 (Dense)              (None, 1)                 513     

Answer 1

从Keras文档中，您可以看到填充是由default=valid填充的，因此没有填充，跨距大小为1。然后输出形状明显是148 x 148。

要计算这个值，可以使用以下公式：

O = (W - K + 2P)/S + 1

其中O是输出高度/宽度，W是输入高度/宽度，K是滤波器尺寸，P是填充，S是步长。

关于第二个参数，您有一个16的特征映射，内核大小为3x3，因此您有16x(3x3)，即144。然后，你有三个颜色通道，使144 x3= 432，然后你需要添加16个偏差，这使448;)希望这有帮助！