学习pytorch9 神经网络-卷积层

神经网络-卷积层

官网

https://pytorch.org/docs/stable/nn.html#convolution-layers

图像识别常用conv2d 二维卷积 nn.Conv2d
https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d

卷积数据公式

参数说明

Parameters:

in_channels (int) -- Number of channels in the input image 输入通道
out_channels (int) -- Number of channels produced by the convolution 输出通道
kernel_size (int or tuple) -- Size of the convolving kernel 卷积核大小
stride (int or tuple, optional) -- Stride of the convolution. Default: 1 每次卷积走多少步，横向纵向的步径大小
padding (int, tuple or str, optional) -- Padding added to all four sides of the input. Default: 0 是否在卷积过程中对输入图像的边缘进行填充
padding_mode (str, optional) -- 'zeros', 'reflect', 'replicate' or 'circular'. Default: 'zeros' 填充数据的模式是什么，默认为zeros，填充的都是0
dilation (int or tuple, optional) -- Spacing between kernel elements. Default: 1 卷积核中间的距离? 一般不改不常用
groups (int, optional) -- Number of blocked connections from input channels to output channels. Default: 1 一般不改不常用
bias (bool, optional) -- If True, adds a learnable bias to the output. Default: True 添加偏置值，默认为True添加偏置值

卷积运算演示

https://github.com/vdumoulin/conv_arithmetic/blob/master/README.md

绿色方格：表示输出图像

蓝色方格：表示输入图像

蓝色方格中的深色阴影部分：表示kernel 卷积核

白色虚线：表示padding填充

动画中深色阴影上下左右整体移动的方格数，表示stride的大小

输入输出channel

两个卷积核做两次卷积，叠加输出一起是out_channel=2

代码

注意点：

super()括号里面没有内容，自动填充的self应该去掉

py 复制代码

super().__init__()

code

py 复制代码

import torch
import torchvision
from torch import nn
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter

test_set = torchvision.datasets.CIFAR10(root='./dataset', train=False, transform=torchvision.transforms.ToTensor(), download=True)

dataloader = DataLoader(test_set, batch_size=64, shuffle=False)

class NnConv2d(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(in_channels=3, out_channels=6, kernel_size=3, stride=1, padding=0)

    def forward(self, x):
        x = self.conv1(x)
        return x

nnconvd = NnConv2d()
writer = SummaryWriter('./logs')
step = 0
for data in dataloader:
    imgs, targets = data
    output = nnconvd(imgs)
    print(imgs.shape)
    print(output.shape)
    writer.add_images("input", imgs)
    output = output.reshape([-1, 3, 30, 30])
    writer.add_images("output", output)
    step += 1

执行结果

sh 复制代码

......
torch.Size([64, 6, 30, 30])
torch.Size([64, 3, 32, 32])
torch.Size([64, 6, 30, 30])
torch.Size([64, 3, 32, 32])
torch.Size([64, 6, 30, 30])
torch.Size([64, 3, 32, 32])
torch.Size([64, 6, 30, 30])
torch.Size([64, 3, 32, 32])
torch.Size([64, 6, 30, 30])
torch.Size([64, 3, 32, 32])
torch.Size([64, 6, 30, 30])
torch.Size([16, 3, 32, 32])
torch.Size([16, 6, 30, 30])  # 最后一个batch16是因为drop_last默认为False，最后没除尽的也要保留，参与训练