Introduction notes to pytorch deep learning (10) neural network convolution layer

Course notes , Course link
The learning notes are synchronously posted on my Personal website On , Welcome to check .

One 、torch.nn.Conv2d brief introduction

Open the official website file torch.nn.Conv2d Section

You can see some of the parameters required for this function , The first five parameters are commonly used , The specific introduction is shown in the figure below ：

Color pictures are generally RGB Format , A pixel has RGB Three parameters , therefore channels by 3.
official A dynamic diagram of convolution operation is provided ：

Yes padding Of , Yes strides Of .

• kernel_size Is used to set the size of convolution kernel , It can be int perhaps tuple data type , For example, set to 3 when , The size of the convolution kernel is 3*3,. In the actual training process , The network will constantly adjust the convolution kernel .
• out-channels Is the number of channels of the output image , When the number of input picture channels is n when , Will use n Channel convolution kernel to convolute the input image , If only one n Channel convolution kernel , The number of channels generated is 1 Output , If you have any m individual n Channel convolution kernel , Then the number of output channels is m.

Two 、 Code demonstration

Sample code ：

import torch
from torch import nn
import torchvision
from torch.nn import Conv2d
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter

dataset = torchvision.datasets.CIFAR10("./dataset", train=False, transform=torchvision.transforms.ToTensor(),
                                       download=True)
dataloader = DataLoader(dataset, batch_size=64)


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

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


Net1 = Net()

writer = SummaryWriter("./logs")
step = 0
for data in dataloader:
    imgs, targets = data
    output = Net1(imgs)
    # print(imgs.shape)
    # print(output.shape)
    # torch.Size([64, 3, 32, 32])
    writer.add_images("input", imgs, step)
    # torch.Size([64, 6, 30, 30])
    output = torch.reshape(output, (-1, 3, 30, 30))
    writer.add_images("output", output, step)
    step = step + 1

use tensorboard To view the ：
Input ：

Output ：

3、 ... and 、VGG16 CNN Model brief introduction

So let's see Vgg16 This CNN Model

2242243 The first and second operation of the input image , It's all convoluted 22422464 Result , The output channel is 64, Constant size , Explain what is going on padding,padding The size can be deduced by the formula ：