Feature Map尺寸计算
Feature Map尺寸计算
-
普通卷积操作
H i n H_{in} Hin 输入图像Height
W i n W_{in} Win 输入图像Width
p a d d i n g padding padding 填充大小(图像四周都添加相同的 p a d d i n g padding padding)
k e r n e l _ s i z e kernel\_size kernel_size 卷积核大小(假设卷积核宽高一样)
s t r i d e stride stride 卷积核移动步长
H o u t H_{out} Hout 输出图像Height
W o u t W_{out} Wout 输出图像Width
H o u t = ⌊ H i n + 2 ∗ p a d d i n g − k e r n e l _ s i z e s t r i d e ⌋ H_{out} = \lfloor \frac{H_{in} + 2 * padding - kernel\_size}{stride} \rfloor Hout=⌊strideHin+2∗padding−kernel_size⌋
W o u t = ⌊ W i n + 2 ∗ p a d d i n g − k e r n e l _ s i z e s t r i d e ⌋ W_{out} = \lfloor \frac{W_{in} + 2 * padding - kernel\_size}{stride} \rfloor Wout=⌊strideWin+2∗padding−kernel_size⌋
卷积层的输入尺寸计算
经过卷积运算后知道特征图大小,反推卷积层输入图像尺寸
H i n H_{in} Hin 卷积层输入图像Height
W i n W_{in} Win 卷积层输入图像Width
p a d d i n g padding padding 填充大小(图像四周都添加相同的 p a d d i n g padding padding)
k e r n e l _ s i z e kernel\_size kernel_size 卷积核大小(假设卷积核宽高一样)
s t r i d e stride stride 卷积核移动步长
H o u t H_{out} Hout 卷积层输出图像Height
W
o
u
t
W_{out}
Wout 卷积层输出图像Width
H
i
n
=
s
t
r
i
d
e
∗
(
H
o
u
t
−
1
)
+
k
e
r
n
e
l
_
s
i
z
e
−
2
∗
p
a
d
d
i
n
g
H_{in} = stride*(H_{out}-1)+kernel\_size-2*padding
Hin=stride∗(Hout−1)+kernel_size−2∗padding
W i n = s t r i d e ∗ ( W o u t − 1 ) + k e r n e l _ s i z e − 2 ∗ p a d d i n g W_{in} = stride*(W_{out}-1)+kernel\_size-2*padding Win=stride∗(Wout−1)+kernel_size−2∗padding