網(wǎng)站首頁編程語言正文

pytorch中的torch.nn.Conv2d()函數(shù)圖文詳解_python

作者：夏普通 ? 更新時(shí)間： 2022-05-01 編程語言

官網(wǎng)

nn.Conv2d：對(duì)由多個(gè)輸入平面組成的輸入信號(hào)進(jìn)行二維卷積

參數(shù)詳解

torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True)

參數(shù)	參數(shù)類型	?	?
in_channels	int	Number of channels in the input image	輸入圖像通道數(shù)
out_channels	int	Number of channels produced by the convolution	卷積產(chǎn)生的通道數(shù)
kernel_size	(int or tuple)	Size of the convolving kernel	卷積核尺寸，可以設(shè)為1個(gè)int型數(shù)或者一個(gè)(int, int)型的元組。例如(2,3)是高2寬3卷積核
stride	(int or tuple, optional)	Stride of the convolution. Default: 1	卷積步長，默認(rèn)為1。可以設(shè)為1個(gè)int型數(shù)或者一個(gè)(int, int)型的元組。
padding	(int or tuple, optional)	Zero-padding added to both sides of the input. Default: 0	填充操作，控制padding_mode的數(shù)目。
padding_mode	(string, optional)	‘zeros’, ‘reflect’, ‘replicate’ or ‘circular’. Default: ‘zeros’	padding模式，默認(rèn)為Zero-padding 。
dilation	(int or tuple, optional)	Spacing between kernel elements. Default: 1	擴(kuò)張操作：控制kernel點(diǎn)（卷積核點(diǎn)）的間距，默認(rèn)值:1。
groups	(int, optional)	Number of blocked connections from input channels to output channels. Default: 1	group參數(shù)的作用是控制分組卷積，默認(rèn)不分組，為1組。
bias	(bool, optional)	If True, adds a learnable bias to the output. Default: True	為真，則在輸出中添加一個(gè)可學(xué)習(xí)的偏差。默認(rèn)：True。

dilation操作動(dòng)圖演示如下：

Dilated Convolution with a 3 x 3 kernel and dilation rate 2

擴(kuò)張卷積核為3×3，擴(kuò)張率為2

Group Convolution顧名思義，則是對(duì)輸入feature map進(jìn)行分組，然后每組分別卷積。

三、代碼實(shí)例

import torch

x = torch.randn(3,1,5,4)
print(x)

conv = torch.nn.Conv2d(1,4,(2,3))
res = conv(x)

print(res.shape)    # torch.Size([3, 4, 4, 2])

輸入：x[ batch_size, channels, height_1, width_1 ]

卷積操作：Conv2d[ channels, output, height_2, width_2 ]

輸出：res[ batch_size,output, height_3, width_3 ]

一個(gè)樣本卷積示例：

原文鏈接：https://blog.csdn.net/qq_34243930/article/details/107231539