5.1-5.2 卷积神经网络——卷积和池化
卷积神经网络——卷积和池化
Convolution Layer
- 可以保全输入的空间结构
- 卷积神经网络越深所学习到的特征越高阶
- 卷积层输出大小公式:
s
i
z
e
=
(
N
+
2
P
−
F
)
/
s
t
r
i
d
e
+
1
size = (N + 2P - F) / stride + 1
size=(N+2P−F)/stride+1 -
1
×
1
1 imes 1
1×1 convolution layers make perfect sense
视觉之外的卷积神经网络
-
5
×
5
5 imes 5
5×5 filters ->
5
×
5
5 imes 5
5×5 receptive field for each neuron
Pooling layer
- make the representations smaller and more manageable
- operates over each activation map independently
- Note that it is not common to use zero-padding for Pooling layers
原文地址:https://www.cnblogs.com/lsl1229840757/p/14122583.html