[Pytorch] 卷积尺寸计算

公式来自官方文档，戳->（Conv3d — PyTorch master documentation）
本文仅作记录，顺便练习Latex语法

2D

(H_{out}=frac{H_{in}+2 imes padding[0]-dilation[0] imes(kernel\_size[0]-1)-1}{stride[1]}+1)
(W_{out}=frac{W_{in}+2 imes padding[1]-dilation[1] imes(kernel\_size[1]-1)-1}{stride[2]}+1)

如果默认dilation=1的话：

(H_{out}=frac{H_{in}+2 imes padding[0]-kernel\_size[0]}{stride[1]}+1)
(W_{out}=frac{W_{in}+2 imes padding[1]-kernel\_size[1]}{stride[2]}+1)

如果默认dilation=1, stride=1的话：

(H_{out}=H_{in}+2 imes padding[0]-kernel\_size[0]+1)
(W_{out}=W_{in}+2 imes padding[1]-kernel\_size[1]+1)

如果默认dilation=1, stride=1, padding=0的话：

(H_{out}=H_{in}-kernel\_size[0]+1)
(W_{out}=W_{in}-kernel\_size[1]+1)

如果默认dilation=1, padding=0的话：

(H_{out}=frac{H_{in}-kernel\_size[0]}{stride[1]}+1)
(W_{out}=frac{W_{in}-kernel\_size[1]}{stride[2]}+1)

3D

(D_{out}=frac{D_{in}+2 imes padding[0]-dilation[0] imes(kernel\_size[0]-1)-1}{stride[0]}+1)
(H_{out}=frac{H_{in}+2 imes padding[1]-dilation[1] imes(kernel\_size[1]-1)-1}{stride[1]}+1)
(W_{out}=frac{W_{in}+2 imes padding[2]-dilation[2] imes(kernel\_size[2]-1)-1}{stride[2]}+1)

如果默认dilation=1的话：

(D_{out}=frac{D_{in}+2 imes padding[0]-kernel\_size[0]}{stride[0]}+1)
(H_{out}=frac{H_{in}+2 imes padding[1]-kernel\_size[1]}{stride[1]}+1)
(W_{out}=frac{W_{in}+2 imes padding[2]-kernel\_size[2]}{stride[2]}+1)

如果默认dilation=1, stride=1的话：

(D_{out}=D_{in}+2 imes padding[0]-kernel\_size[0]+1)
(H_{out}=H_{in}+2 imes padding[1]-kernel\_size[1]+1)
(W_{out}=W_{in}+2 imes padding[2]-kernel\_size[2]+1)

如果默认dilation=1, stride=1, padding=0的话：
(D_{out}=D_{in}-kernel\_size[0]+1)
(H_{out}=H_{in}-kernel\_size[1]+1)
(W_{out}=W_{in}-kernel\_size[2]+1)

如果默认dilation=1, padding=0的话：

(D_{out}=frac{D_{in}-kernel\_size[0]}{stride[1]}+1)
(H_{out}=frac{H_{in}-kernel\_size[1]}{stride[1]}+1)
(W_{out}=frac{W_{in}-kernel\_size[2]}{stride[2]}+1)

附用法

Parameters
in_channels (int) – Number of channels in the input image
out_channels (int) – Number of channels produced by the convolution
kernel_size (int or tuple) – Size of the convolving kernel
stride (int or tuple, optional) – Stride of the convolution. Default: 1
padding (int or tuple, optional) – Zero-padding added to both sides of the input. Default: 0
padding_mode (string, optional) – 'zeros', 'reflect', 'replicate' or 'circular'. Default: 'zeros'
dilation (int or tuple, optional) – Spacing between kernel elements. Default: 1
groups (int, optional) – Number of blocked connections from input channels to output channels. Default: 1
bias (bool, optional) – If True, adds a learnable bias to the output. Default: True