CIFAR10神经网络初步搭建

ZilliaxBergling

已于 2024-08-13 15:36:14 修改

阅读量201

点赞数 2

分类专栏： Pytorch 文章标签： pytorch 人工智能机器学习深度学习

于 2024-08-13 10:29:51 首次发布

本文链接：https://blog.csdn.net/RaltBergling/article/details/141151568

版权

Pytorch 专栏收录该内容

17 篇文章 0 订阅

订阅专栏

这是网络上一个的示例模型，先按照这个模型搭建；

首先是Conv2d需要的公式

卷积核里的padding和stride需要自己计算出来，通过公式：

输出高度 = [输入高度 + 2 * padding - dialaction * (kernel_size[0] - 1)] / stride + 1

比如本次CIFAR10的输入RGB图片就是3通道，32 * 32格式，卷积核设置为5 * 5，输出图片为32通道，32 * 32，则带入可得
32 = [32 + 2 * padding - 1 * (5 - 1)]/stride + 1
31 = [27 + 2*padding] / stride
得到最优解为padding = 2 , stride = 1;

import torch
from torch import nn
from torch.nn import Conv2d, MaxPool2d, Flatten, Linear
from torch.utils.tensorboard import SummaryWriter

"""
卷积核里的padding和stride需要自己计算出来，通过公式：

    输出高度 = [输入高度 + 2 * padding - dialaction * (kernel_size[0] - 1)] / stride + 1

比如本次CIFAR10的输入图片就是3通道，32 * 32格式，卷积核设置为5 * 5，输出图片为32通道，32 * 32，则带入可得
    32 = [32 + 2 * padding - 1 * (5 - 1)]/stride + 1
    31 = [27 + 2*padding] / stride
得到最优解为padding = 2 , stride = 1;

Conv2d(in_channel , outchannel , kernel_size , stride , padding)
MaxPool2d(kernel_size , stride , padding , dialaction)
Flatten()
Linear(input_feature , output_feature)
"""

class Zilliax(nn.Module):
    def __init__(self):
        super(Zilliax, self).__init__()
        self.conv1 = Conv2d(3, 32, 5, stride=1, padding=2)  # 32 = [32 + 2 * padding - 1 * (5 - 1)]/stride + 1
        self.maxpool1 = MaxPool2d(2)  # 池化核的strde默认和size一样
        self.conv2 = Conv2d(32, 32, 5, padding=2)  # 32 = [32 + 2 * padding - 1 * (5 - 1)]/stride + 1
        self.maxpool2 = MaxPool2d(2)
        self.conv3 = Conv2d(32, 64, 5, padding=2)
        self.maxpool3 = MaxPool2d(2)
        self.flatten = Flatten()
        self.linear1 = Linear(1024, 64)  #input = 64 * 4 * 4 = 1024 * 1 * 1
        self.linear2 = Linear(64, 10)

    def forward(self, x):
        x = self.conv1(x)
        x = self.maxpool1(x)
        x = self.conv2(x)
        x = self.maxpool2(x)
        x = self.conv3(x)
        x = self.maxpool3(x)
        x = self.flatten(x)
        # 不要这两个linear可以得到linear1里的1024
        x = self.linear1(x)
        x = self.linear2(x)
        return x

z = Zilliax()
print(z)

"""
输出得到整个神经网络的结构：
Conv2d(3, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
Conv2d(32, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
Conv2d(32, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
Flatten(start_dim=1, end_dim=-1)
Linear(in_features=1024, out_features=64, bias=True)
Linear(in_features=64, out_features=10, bias=True)
"""

# 用ones试错神经网络中linear1里最开始的那个1024
sample = torch.ones((64, 3, 32, 32))    # 64是batch_size
output = z(sample)
print(output.shape)

# 写入graph，可以看到很多值和整个模型运行的过程
writer = SummaryWriter('./logs')
writer.add_graph(z,sample)
writer.close()

很显然，这样的神经网络写法非常麻烦，所以可以使用Sequential对整个过程进行排序和打包：

import torch
from torch import nn
from torch.nn import Conv2d, MaxPool2d, Flatten, Linear
from torch.utils.tensorboard import SummaryWriter

"""
卷积核里的padding和stride需要自己计算出来，通过公式：

    输出高度 = [输入高度 + 2 * padding - dialaction * (kernel_size[0] - 1)] / stride + 1

比如本次CIFAR10的输入图片就是3通道，32 * 32格式，卷积核设置为5 * 5，输出图片为32通道，32 * 32，则带入可得
    32 = [32 + 2 * padding - 1 * (5 - 1)]/stride + 1
    31 = [27 + 2*padding] / stride
得到最优解为padding = 2 , stride = 1;

Conv2d(in_channel , outchannel , kernel_size , stride , padding)
MaxPool2d(kernel_size , stride , padding , dialaction)
Flatten()
Linear(input_feature , output_feature)
"""


class Zilliax(nn.Module):
    def __init__(self):
        super(Zilliax, self).__init__()
        """
        self.conv1 = Conv2d(3, 32, 5, stride=1, padding=2)  # 32 = [32 + 2 * padding - 1 * (5 - 1)]/stride + 1
        self.maxpool1 = MaxPool2d(2)  # 池化核的strde默认和size一样
        self.conv2 = Conv2d(32, 32, 5, padding=2)  # 32 = [32 + 2 * padding - 1 * (5 - 1)]/stride + 1
        self.maxpool2 = MaxPool2d(2)
        self.conv3 = Conv2d(32, 64, 5, padding=2)
        self.maxpool3 = MaxPool2d(2)
        self.flatten = Flatten()
        self.linear1 = Linear(1024, 64)  #input = 64 * 4 * 4 = 1024 * 1 * 1
        self.linear2 = Linear(64, 10)
        """
        # sequential简化模型
        self.module = nn.Sequential(
            Conv2d(3, 32, 5, stride=1, padding=2),
            MaxPool2d(2, stride=2),
            Conv2d(32, 32, 5, stride=1, padding=2),
            MaxPool2d(2, stride=2),
            Conv2d(32, 64, 5, stride=1, padding=2),
            MaxPool2d(2, stride=2),
            Flatten(),
            Linear(1024, 64),
            Linear(64, 10)
        )


    def forward(self, x):
        """
        x = self.conv1(x)
        x = self.maxpool1(x)
        x = self.conv2(x)
        x = self.maxpool2(x)
        x = self.conv3(x)
        x = self.maxpool3(x)
        x = self.flatten(x)
        # 不要这两个linear可以得到linear1里的1024
        x = self.linear1(x)
        x = self.linear2(x)
        """
        x = self.module(x)
        return x


z = Zilliax()
print(z)

# 用ones试错神经网络中linear1里最开始的那个1024
sample = torch.ones((64, 3, 32, 32))    # 64是batch_size
output = z(sample)
print(output.shape)

# 写入graph，可以看到很多值和整个模型运行的过程
writer = SummaryWriter('./logs')
writer.add_graph(z,sample)
writer.close()