神经网络——卷积层 小土堆 P18

1. nn.conv2d的使用

2.用CIFAR作为数据集输入

3.卷积层的设置

4.读取dataloader中的数据，并进行卷积，查看卷积前和卷积后的变化

5.用tensorboard读取卷积后的图像

结果展示

完整代码

---

卷积层

1. nn.conv2d的使用

红色为常用的

in_channels:输入的通道数
out_channels:输出的通道数
kernel_size:卷积核大小
stride: 卷积过程中步径大小
padding:填充
dilation
bias：偏置
 

2.用CIFAR作为数据集输入

# 数据集创建
dataset = torchvision.datasets.CIFAR10(root="dataset", train=False, transform=torchvision.transforms.ToTensor())
# 加载数据，64个图片为一个包
dataloader = DataLoader(dataset, batch_size=64)

3.卷积层的设置

class M(nn.Module):
    def __init__(self):
        super().__init__()  # 父类初始化
        self.conv1 = Conv2d(in_channels=3, out_channels=6, kernel_size=3, stride=1, padding=0)

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

4.读取dataloader中的数据，并进行卷积，查看卷积前和卷积后的变化

conv = M()  # 实例化
for data in dataloader:
    imgs, targets = data
    output = conv(imgs)
    print(imgs.shape)
    print(output.shape)

可以看到由三通道变为六通道，

像素32*32由于卷积的影响变成30*30

5.用tensorboard读取卷积后的图像

writer = SummaryWriter("logs")

step = 0
for data in dataloader:
    imgs, targets = data
    output = conv(imgs)
    print(imgs.shape)
    print(output.shape)

    # torch.Size([64, 3, 32, 32])
    writer.add_images(tag="image_input", img_tensor=imgs, global_step=step)
    # torch.Size([64, 6, 30, 30]) -> [XXX, 3, 30, 30] 六个通道变为三个通道
    output = torch.reshape(input=output, shape=(-1, 3, 30, 30))
    writer.add_images(tag="image_conv", img_tensor=output, global_step=step)
    step = step + 1

writer.close()

结果展示

完整代码

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

# 数据集创建
dataset = torchvision.datasets.CIFAR10(root="dataset", train=False, transform=torchvision.transforms.ToTensor())
# 加载数据，64个图片为一个包
dataloader = DataLoader(dataset, batch_size=64)


class M(nn.Module):
    def __init__(self):
        super().__init__()  # 父类初始化
        self.conv1 = Conv2d(in_channels=3, out_channels=6, kernel_size=3, stride=1, padding=0)

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


conv = M()  # 实例化
writer = SummaryWriter("logs")

step = 0
for data in dataloader:
    imgs, targets = data
    output = conv(imgs)
    print(imgs.shape)
    print(output.shape)

    # torch.Size([64, 3, 32, 32])
    writer.add_images(tag="image_input", img_tensor=imgs, global_step=step)
    # torch.Size([64, 6, 30, 30]) -> [XXX, 3, 30, 30] 六个通道变为三个通道
    output = torch.reshape(input=output, shape=(-1, 3, 30, 30))
    writer.add_images(tag="image_conv", img_tensor=output, global_step=step)
    step = step + 1

writer.close()