---

---

前言

使用 B站小土堆课程的笔记

---

一、定义概念 + 缩写

• 卷积层是神经网络中用于突出特征来进行分类任务的层。

二、性质

• 卷积核例子：vgg16 model
  

三、代码

• 添加库

python代码块import os
import torch
import torchvision # torchvision 通常用于计算机视觉任务
from torch import nn
from torch.nn import Conv2d
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter

• 下载/加载数据集，加载数据

# 这是一个类，用于下载和加载 CIFAR-10 数据集。CIFAR-10 是一个常用的小型图像数据集，用于训练机器学习和计算机视觉算法。它包含10个类别，每个类别有6000张32x32的彩色图像，总共有60000张图像。
# transform=torchvision.transforms.ToTensor() 转换格式
dataset = torchvision.datasets.CIFAR10("../data", train=False, transform=torchvision.transforms.ToTensor(),
                                       download=True)

# load data
dataloader = DataLoader(dataset, batch_size=64)

• 定义类，搭建简单神经网络

# 搭建简单 NN
class Tudui(nn.Module):
    def __init__(self):

        # 继承
        super(Tudui, self).__init__()

        # init
        self.conv1 = Conv2d(in_channels=3, out_channels=6, kernel_size=3, stride=1, padding=0)

    # 卷积层
    def forward(self, x):
        x = self.conv1(x) # conv
        return x
        
# create an example
tudui = Tudui()

• 加载到 tensorboard

# log writer, write to tensorboard
writer = SummaryWriter("../logs")

• 进行卷积

# init counter
step = 0

# 循环进行卷积操作
for data in dataloader:
    imgs, targets = data
    output = tudui(imgs) # def a output
    print(imgs.shape)
    print(output.shape)
    # torch.Size([64, 3, 32, 32])

    # add img to tensorboard
    writer.add_images("input", imgs, step)
    # torch.Size([64, 6, 30, 30])  -> [xxx, 3, 30, 30]

    output = torch.reshape(output, (-1, 3, 30, 30))
    writer.add_images("output", output, step)

    # counter
    step = step + 1

• 直接在 py 中加载 tensorboard，端口=6006；关闭 writer

# open tensorboard
os.system('tensorboard --logdir=logs --port=6006')
# tensorboard --logdir="D:\Libraries\projects\python\001 learningTest and small task\a005_pytorch\a002_lesson_src\logs" --port=6006

writer.close()

---

总结

---

参考文献

[1]