当前位置：首页 > article >正文

深度学习项目--基于ResNet和DenseNet混合架构网络论文的复现(pytorch实现)

article 2025/3/22 5:24:49

🍨 本文为🔗365天深度学习训练营中的学习记录博客
🍖 原作者：K同学啊

前言

如果说最经典的神经网络，ResNet当之无愧，后面基于ResNet网络提出DenseNet网络，也获得了best paper，本文是对这两个网络进行结合的一次论文复现，并用其进行了实验；
ResNet讲解： https://blog.csdn.net/weixin_74085818/article/details/145786990?spm=1001.2014.3001.5501
DenseNet讲解：https://blog.csdn.net/weixin_74085818/article/details/146102290?spm=1001.2014.3001.5501
欢迎收藏 + 关注，本人将会持续更新

文章目录

1、ResNet网络与DenseNet网络模型探索
2、神经网络搭建与训练
1、导入数据
1、导入库
2、查看数据信息和导入数据
3、展示数据
4、数据导入
5、数据划分
6、动态加载数据

2、构建融合网络
3、模型训练
1、构建训练集
2、构建测试集
3、设置超参数

4、模型训练
5、结果可视化
6、模型评估

1、ResNet网络与DenseNet网络模型探索

常见的一些模型融合方法有：

特征级融合：可以在两个网络的某个中间层进行特征图的直接拼接(concatenation)或元素级相加(addition)。
决策级融合：另一种方法是在网络的输出端进行融合，即先分别使用ResNet和DenseNet对输入数据进行处理
混合架构设计：还可以设计一种新的网络架构，将ResNet和DenseNet的特点结合起来。例如，在一个网络中
基于注意力机制的融合：可以采用注意力机制动态地融合来自ResNet和DenseNet的信息。

本文采用的是第三种，参考论文：论文

Resnet模型和DenseNet模型特点：

ResNet：通过建立前面层与后面层之间的“短路连接”（shortcu），其特征则直接进行sum操作，因此channel数不变；
DenseNet：通过建立的是前面所有层与后面层的紧密连接（dense connection），其特征在channel维度上的直接concat来实现特征重用（feature reuse），因此channel数增加；

在《论文》中，作者发现：ResNet更侧重于特征的复用，而DenseNet则更侧重于特征的生成。

故作者提出了DPN网络，如图所示：

在这里插入图片描述

这个图比较难看懂，换个更清晰的图看：

在这里插入图片描述

这个图就容易看懂多了，DPN网络核心的就是上图中，蓝框和红框的东西，在DPN中，训练模块Block中，将输出的信息进行分拆，然后又进行融合，分拆和融合的思想是，ResNet的特征复用，与DenseNet的创建新特征：

混合模型融合分为两步: 思想是resnet模型的特征复用，densenet模型的创建新特征(dense_channel)
- ResNet思想，特征复用，这里同时结合densenet思路，在通道进行融合；
- 创建新特征：分别结合残差连接的resnet网络模块，与没有使用残差连接的网络模块数据，进行通道融合；
- 新特征通道数量 = out_channel + 2 * dense_channel。
- 具体实现，看代码即可：

参考ResNet网络，可以发现，这个基本模块是基于ResNet模型进行改进的。

在这里插入图片描述

在ResNet网络中，采用1*1，3*3，1*1的三层网络进行特征提取，该论文中也是采用这个结构，但是不同的是第二层3 * 3还加入了分组卷积的方式，以便更好的提取特征，具体看代码即可，代码注释详细。

2、神经网络搭建与训练

1、导入数据

1、导入库

import torch  
import torch.nn as nn
import torchvision 
import numpy as np 
import os, PIL, pathlib 
from collections import OrderedDict
import re
from torch.hub import load_state_dict_from_url

# 设置设备
device = "cuda" if torch.cuda.is_available() else "cpu"

device

'cuda'

2、查看数据信息和导入数据

data_dir = "./data/"

data_dir = pathlib.Path(data_dir)

# 类别数量
classnames = [str(path).split("\\")[0] for path in os.listdir(data_dir)]

classnames

['Bananaquit', 'Black Skimmer', 'Black Throated Bushtiti', 'Cockatoo']

3、展示数据

import matplotlib.pylab as plt  
from PIL import Image 

# 获取文件名称
data_path_name = "./data/Cockatoo/"  # 不患病的
data_path_list = [f for f in os.listdir(data_path_name) if f.endswith(('jpg', 'png'))]

# 创建画板
fig, axes = plt.subplots(2, 8, figsize=(16, 6))

for ax, img_file in zip(axes.flat, data_path_list):
    path_name = os.path.join(data_path_name, img_file)
    img = Image.open(path_name) # 打开
    # 显示
    ax.imshow(img)
    ax.axis('off')
    
plt.show()

在这里插入图片描述

4、数据导入

from torchvision import transforms, datasets 

# 数据统一格式
img_height = 224
img_width = 224 

data_tranforms = transforms.Compose([
    transforms.Resize([img_height, img_width]),
    transforms.ToTensor(),
    transforms.Normalize(   # 归一化
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225] 
    )
])

# 加载所有数据
total_data = datasets.ImageFolder(root="./data/", transform=data_tranforms)

5、数据划分

# 大小 8 : 2
train_size = int(len(total_data) * 0.8)
test_size = len(total_data) - train_size 

train_data, test_data = torch.utils.data.random_split(total_data, [train_size, test_size])

6、动态加载数据

batch_size = 32

train_dl = torch.utils.data.DataLoader(
    train_data,
    batch_size=batch_size,
    shuffle=True
)

test_dl = torch.utils.data.DataLoader(
    test_data,
    batch_size=batch_size,
    shuffle=False
)

# 查看数据维度
for data, labels in train_dl:
    print("data shape[N, C, H, W]: ", data.shape)
    print("labels: ", labels)
    break

data shape[N, C, H, W]:  torch.Size([32, 3, 224, 224])
labels:  tensor([2, 2, 3, 0, 0, 0, 0, 1, 0, 1, 2, 0, 0, 3, 0, 1, 0, 1, 3, 3, 2, 0, 0, 3,
        3, 3, 3, 1, 3, 3, 1, 2])

2、构建融合网络

ResNet和DenseNet 神经网络是以，ResNet为基础，故在特征提取中，采用和ResNet网络一样的1 * 1、3 * 3、1 * 1卷积核.

ResNet和DenseNet结合网络图为：
在这里插入图片描述

import torch.nn.functional as F 

# DPN模块：Block，结合网络图即可
class Block(nn.Module):
    '''  
    in_channel: 输入通道数
    mid_channel: 中间通道数
    out_channel: ResNet输出的通道数
    dense_channel: DenseNet网络产生新的特征通道
    groups: 分组卷积参数
    is_shortcut: ResNet是否进行残差操作
    '''
    def __init__(self, in_channel, mid_channel, out_channel, dense_channel, stride, groups, is_shortcut=False):
        super().__init__()
        
        # 参数存储
        self.is_shortcut = is_shortcut
        self.out_channel = out_channel
        self.dense_channel = dense_channel
        
        # 构建卷积模块，三层（Conv2d、BN、ReLU）
        self.conv1 = nn.Sequential(
            # 卷积核 1 * 1
            nn.Conv2d(in_channel, mid_channel, kernel_size=1, bias=False),  # 偏置设置为Fasle，因为下一层是BN层，BN本身启用了bias
            nn.BatchNorm2d(mid_channel),
            nn.ReLU()
        )
        
        self.conv2 = nn.Sequential(
            # 分组卷积（为了提取更多特征），且卷积核为3 * 3
            nn.Conv2d(mid_channel, mid_channel, kernel_size=3, stride=stride, groups=groups, padding=1, bias=False),
            nn.BatchNorm2d(mid_channel),
            nn.ReLU()
        )
        
        self.conv3 = nn.Sequential(
            # 卷积核 1 * 1，维持维度
            nn.Conv2d(mid_channel, out_channel + dense_channel, kernel_size=1, bias=False),
            nn.BatchNorm2d(out_channel + dense_channel)
        )
        
        # 是否启动ResNet的残差连接, 这个对应上面网络图中间那一模块
        if self.is_shortcut:
            self.shortcut = nn.Sequential(
                nn.Conv2d(in_channel, out_channel + dense_channel, kernel_size=3, stride=stride, padding=1,bias=False),
                nn.BatchNorm2d(out_channel + dense_channel)
            )
            
        self.relu = nn.ReLU(inplace=True)
        
    def forward(self, x):
        a = x
        x = self.conv1(x)
        x = self.conv2(x)
        x = self.conv3(x)
        if self.is_shortcut:
            a = self.shortcut(a)
        
        # 核心： 混合模型融合
        '''  
        混合模型融合分为两步: 思想是resnet模型的特征复用，densenet模型的创建新特征(dense_channel)
            1、ResNet思想，特征复用，这里同时结合densenet思路，在用到通道进行融合
            2、创建新特征：分别结合残差连接的resnet网络模块，与没有使用残差连接的网络模块数据，进行通道融合
        新特征通道数量 = out_channel + 2 * dense_channel
        '''
        x = torch.cat([a[:, :self.out_channel, :, :]+x[:, :self.out_channel, :, :], a[:, self.out_channel:, :, :], x[:, self.out_channel:, :, :]], dim=1)
        
        return x 
    
class DPN(nn.Module):
    # cfg：参数字典
    def __init__(self, cfg):
        super().__init__()
        
        # 储存参数
        self.group = cfg['group'] # 有一层为分组卷积
        self.in_channel = cfg['in_channels']
        mid_channel = cfg['mid_channels']
        out_channel = cfg['out_channels']
        dense_channel = cfg['dense_channels']
        num = cfg['num']   # Blcok数量，元组
        
        # 数据处理模块，即网络头
        self.conv1 = nn.Sequential(
            nn.Conv2d(3, self.in_channel, kernel_size=7, padding=3, bias=False, padding_mode='zeros'),  # 卷积核为7，和densenet一样
            nn.BatchNorm2d(self.in_channel),
            nn.ReLU(),
            nn.MaxPool2d(kernel_size=3, stride=2, padding=0)  # 降维
        )
        
        self.conv2 = self._make_layers(mid_channel[0], out_channel[0], dense_channel[0], num[0], stride=1)
        self.conv3 = self._make_layers(mid_channel[1], out_channel[1], dense_channel[1], num[1], stride=2)
        self.conv4 = self._make_layers(mid_channel[2], out_channel[2], dense_channel[2], num[2], stride=2)
        self.conv5 = self._make_layers(mid_channel[3], out_channel[3], dense_channel[3], num[3], stride=2)
        
        self.pool = nn.AdaptiveAvgPool2d((1, 1))
        self.fc = nn.Linear(cfg['out_channels'][3] + (num[3] + 1) * cfg['dense_channels'][3], cfg['classes'])  # Linear层，展开
        
    
    # 构建网络层，从 网络 图像上看，Block的叠加数量不同
    def _make_layers(self, mid_channel, out_channel, dense_channel, num, stride):
        layers = []
        # 每一个部分，都先用shortcut的Block，可以满足浅层特征利用, 增强特征提取作用，提升性能
        layers.append(Block(self.in_channel, mid_channel, out_channel, dense_channel,stride=stride, groups=self.group, is_shortcut=True))
        # 新特征通道数量 = out_channel + 2 * dense_channel
        self.in_channel = out_channel + 2 * dense_channel
        for i in range(1, num):
            layers.append(Block(self.in_channel, mid_channel, out_channel, dense_channel, stride=1, groups=self.group))
            # 每次叠加，通道数量多一倍 dense_channel
            self.in_channel += dense_channel 
        return nn.Sequential(*layers)
    
    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        x = self.conv3(x)
        x = self.conv4(x)
        x = self.conv5(x)
        x = self.pool(x)
        x = torch.flatten(x, start_dim=1)
        x = self.fc(x)
        return x

设置参数与模型构建，在DPN论文中，提供了两种网络结果，DPN92，DPN98

def DPN92(n_class=4):
    cfg = {
        "group" : 32,
        "in_channels" : 64,
        "mid_channels" : (96, 192, 384, 768),
        "out_channels" : (256, 512, 1024, 2048),
        "dense_channels" : (16, 32, 24, 128),
        "num" : (3, 4, 20, 3),
        "classes" : (len(classnames))
    }
    return DPN(cfg)
 
def DPN98(n_class=4):
    cfg = {
        "group" : 40,
        "in_channels" : 96,
        "mid_channels" : (160, 320, 640, 1280),
        "out_channels" : (256, 512, 1024, 2048),
        "dense_channels" : (16, 32, 32, 128),
        "num" : (3, 6, 20, 3),
        "classes" : (len(classnames))
    }
    return DPN(cfg)
 
model = DPN92().to(device)
model

DPN(
  (conv1): Sequential(
    (0): Conv2d(3, 64, kernel_size=(7, 7), stride=(1, 1), padding=(3, 3), bias=False)
    (1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
    (3): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (conv2): Sequential(
    (0): Block(
      (conv1): Sequential(
        (0): Conv2d(64, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(96, 272, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(272, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (shortcut): Sequential(
        (0): Conv2d(64, 272, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (1): BatchNorm2d(272, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (1): Block(
      (conv1): Sequential(
        (0): Conv2d(288, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(96, 272, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(272, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (2): Block(
      (conv1): Sequential(
        (0): Conv2d(304, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(96, 272, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(272, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
  )
  (conv3): Sequential(
    (0): Block(
      (conv1): Sequential(
        (0): Conv2d(320, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(192, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(192, 544, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(544, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (shortcut): Sequential(
        (0): Conv2d(320, 544, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
        (1): BatchNorm2d(544, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (1): Block(
      (conv1): Sequential(
        (0): Conv2d(576, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(192, 544, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(544, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (2): Block(
      (conv1): Sequential(
        (0): Conv2d(608, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(192, 544, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(544, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (3): Block(
      (conv1): Sequential(
        (0): Conv2d(640, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(192, 544, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(544, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
  )
  (conv4): Sequential(
    (0): Block(
      (conv1): Sequential(
        (0): Conv2d(672, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (shortcut): Sequential(
        (0): Conv2d(672, 1048, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (1): Block(
      (conv1): Sequential(
        (0): Conv2d(1072, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (2): Block(
      (conv1): Sequential(
        (0): Conv2d(1096, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (3): Block(
      (conv1): Sequential(
        (0): Conv2d(1120, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (4): Block(
      (conv1): Sequential(
        (0): Conv2d(1144, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (5): Block(
      (conv1): Sequential(
        (0): Conv2d(1168, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (6): Block(
      (conv1): Sequential(
        (0): Conv2d(1192, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (7): Block(
      (conv1): Sequential(
        (0): Conv2d(1216, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (8): Block(
      (conv1): Sequential(
        (0): Conv2d(1240, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (9): Block(
      (conv1): Sequential(
        (0): Conv2d(1264, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (10): Block(
      (conv1): Sequential(
        (0): Conv2d(1288, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (11): Block(
      (conv1): Sequential(
        (0): Conv2d(1312, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (12): Block(
      (conv1): Sequential(
        (0): Conv2d(1336, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (13): Block(
      (conv1): Sequential(
        (0): Conv2d(1360, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (14): Block(
      (conv1): Sequential(
        (0): Conv2d(1384, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (15): Block(
      (conv1): Sequential(
        (0): Conv2d(1408, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (16): Block(
      (conv1): Sequential(
        (0): Conv2d(1432, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (17): Block(
      (conv1): Sequential(
        (0): Conv2d(1456, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (18): Block(
      (conv1): Sequential(
        (0): Conv2d(1480, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (19): Block(
      (conv1): Sequential(
        (0): Conv2d(1504, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(384, 1048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(1048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
  )
  (conv5): Sequential(
    (0): Block(
      (conv1): Sequential(
        (0): Conv2d(1528, 768, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(768, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(768, 768, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(768, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(768, 2176, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(2176, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (shortcut): Sequential(
        (0): Conv2d(1528, 2176, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
        (1): BatchNorm2d(2176, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (1): Block(
      (conv1): Sequential(
        (0): Conv2d(2304, 768, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(768, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(768, 768, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(768, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(768, 2176, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(2176, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
    (2): Block(
      (conv1): Sequential(
        (0): Conv2d(2432, 768, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(768, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv2): Sequential(
        (0): Conv2d(768, 768, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
        (1): BatchNorm2d(768, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (2): ReLU()
      )
      (conv3): Sequential(
        (0): Conv2d(768, 2176, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(2176, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
      (relu): ReLU(inplace=True)
    )
  )
  (pool): AdaptiveAvgPool2d(output_size=(1, 1))
  (fc): Linear(in_features=2560, out_features=4, bias=True)
)

model(torch.randn(32, 3, 224, 224).to(device)).shape

torch.Size([32, 4])

3、模型训练

1、构建训练集

def train(dataloader, model, loss_fn, optimizer):
    size = len(dataloader.dataset)
    batch_size = len(dataloader)
    
    train_acc, train_loss = 0, 0 
    
    for X, y in dataloader:
        X, y = X.to(device), y.to(device)
        
        # 训练
        pred = model(X)
        loss = loss_fn(pred, y)
        
        # 梯度下降法
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        
        # 记录
        train_loss += loss.item()
        train_acc += (pred.argmax(1) == y).type(torch.float).sum().item()
        
    train_acc /= size
    train_loss /= batch_size
    
    return train_acc, train_loss

2、构建测试集

def test(dataloader, model, loss_fn):
    size = len(dataloader.dataset)
    batch_size = len(dataloader)
    
    test_acc, test_loss = 0, 0 
    
    with torch.no_grad():
        for X, y in dataloader:
            X, y = X.to(device), y.to(device)
        
            pred = model(X)
            loss = loss_fn(pred, y)
        
            test_loss += loss.item()
            test_acc += (pred.argmax(1) == y).type(torch.float).sum().item()
        
    test_acc /= size
    test_loss /= batch_size
    
    return test_acc, test_loss

3、设置超参数

loss_fn = nn.CrossEntropyLoss()  # 损失函数     
learn_lr = 1e-4             # 超参数
optimizer = torch.optim.Adam(model.parameters(), lr=learn_lr)   # 优化器

4、模型训练

import copy

train_acc = []
train_loss = []
test_acc = []
test_loss = []

epoches = 40

best_acc = 0    # 设置一个最佳准确率，作为最佳模型的判别指标

for i in range(epoches):
    model.train()
    epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, optimizer)
    
    model.eval()
    epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)

     # 保存最佳模型到 best_model     
    if epoch_test_acc > best_acc:         
        best_acc   = epoch_test_acc         
        best_model = copy.deepcopy(model)
    
    train_acc.append(epoch_train_acc)
    train_loss.append(epoch_train_loss)
    test_acc.append(epoch_test_acc)
    test_loss.append(epoch_test_loss)

    # 获取当前的学习率     
    lr = optimizer.state_dict()['param_groups'][0]['lr']
    
    # 输出
    template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}')
    print(template.format(i + 1, epoch_train_acc*100, epoch_train_loss, epoch_test_acc*100, epoch_test_loss))

PATH = './best_model.pth'  # 保存的参数文件名 
torch.save(best_model.state_dict(), PATH)

print("Done")

Epoch: 1, Train_acc:56.4%, Train_loss:1.281, Test_acc:27.4%, Test_loss:2.379
Epoch: 2, Train_acc:70.6%, Train_loss:0.863, Test_acc:37.2%, Test_loss:2.512
Epoch: 3, Train_acc:79.4%, Train_loss:0.633, Test_acc:59.3%, Test_loss:1.114
Epoch: 4, Train_acc:85.6%, Train_loss:0.446, Test_acc:77.9%, Test_loss:0.714
Epoch: 5, Train_acc:91.4%, Train_loss:0.320, Test_acc:85.8%, Test_loss:0.411
Epoch: 6, Train_acc:90.3%, Train_loss:0.232, Test_acc:92.0%, Test_loss:0.234
Epoch: 7, Train_acc:96.9%, Train_loss:0.111, Test_acc:91.2%, Test_loss:0.196
Epoch: 8, Train_acc:96.7%, Train_loss:0.102, Test_acc:91.2%, Test_loss:0.263
Epoch: 9, Train_acc:96.7%, Train_loss:0.120, Test_acc:88.5%, Test_loss:0.282
Epoch:10, Train_acc:96.0%, Train_loss:0.131, Test_acc:77.9%, Test_loss:1.098
Epoch:11, Train_acc:96.2%, Train_loss:0.198, Test_acc:92.0%, Test_loss:0.319
Epoch:12, Train_acc:94.0%, Train_loss:0.164, Test_acc:90.3%, Test_loss:0.449
Epoch:13, Train_acc:96.2%, Train_loss:0.175, Test_acc:86.7%, Test_loss:0.330
Epoch:14, Train_acc:96.7%, Train_loss:0.085, Test_acc:86.7%, Test_loss:0.495
Epoch:15, Train_acc:98.0%, Train_loss:0.052, Test_acc:84.1%, Test_loss:0.561
Epoch:16, Train_acc:99.1%, Train_loss:0.022, Test_acc:88.5%, Test_loss:0.335
Epoch:17, Train_acc:99.6%, Train_loss:0.021, Test_acc:89.4%, Test_loss:0.272
Epoch:18, Train_acc:99.8%, Train_loss:0.027, Test_acc:88.5%, Test_loss:0.299
Epoch:19, Train_acc:97.1%, Train_loss:0.176, Test_acc:83.2%, Test_loss:0.796
Epoch:20, Train_acc:88.1%, Train_loss:0.588, Test_acc:81.4%, Test_loss:0.775
Epoch:21, Train_acc:87.8%, Train_loss:0.322, Test_acc:54.0%, Test_loss:4.497
Epoch:22, Train_acc:92.9%, Train_loss:0.187, Test_acc:75.2%, Test_loss:1.528
Epoch:23, Train_acc:97.6%, Train_loss:0.081, Test_acc:85.0%, Test_loss:0.750
Epoch:24, Train_acc:97.6%, Train_loss:0.055, Test_acc:91.2%, Test_loss:0.283
Epoch:25, Train_acc:99.6%, Train_loss:0.020, Test_acc:94.7%, Test_loss:0.225
Epoch:26, Train_acc:100.0%, Train_loss:0.007, Test_acc:93.8%, Test_loss:0.267
Epoch:27, Train_acc:99.8%, Train_loss:0.012, Test_acc:90.3%, Test_loss:0.266
Epoch:28, Train_acc:100.0%, Train_loss:0.004, Test_acc:89.4%, Test_loss:0.385
Epoch:29, Train_acc:99.8%, Train_loss:0.047, Test_acc:92.0%, Test_loss:0.224
Epoch:30, Train_acc:98.5%, Train_loss:0.055, Test_acc:82.3%, Test_loss:1.111
Epoch:31, Train_acc:98.7%, Train_loss:0.073, Test_acc:84.1%, Test_loss:0.642
Epoch:32, Train_acc:94.2%, Train_loss:0.159, Test_acc:78.8%, Test_loss:0.788
Epoch:33, Train_acc:96.0%, Train_loss:0.097, Test_acc:89.4%, Test_loss:0.411
Epoch:34, Train_acc:97.8%, Train_loss:0.085, Test_acc:90.3%, Test_loss:0.345
Epoch:35, Train_acc:98.9%, Train_loss:0.039, Test_acc:90.3%, Test_loss:0.633
Epoch:36, Train_acc:99.3%, Train_loss:0.042, Test_acc:85.0%, Test_loss:0.472
Epoch:37, Train_acc:99.3%, Train_loss:0.041, Test_acc:84.1%, Test_loss:0.580
Epoch:38, Train_acc:97.8%, Train_loss:0.161, Test_acc:87.6%, Test_loss:0.526
Epoch:39, Train_acc:97.1%, Train_loss:0.092, Test_acc:88.5%, Test_loss:0.616
Epoch:40, Train_acc:97.3%, Train_loss:0.073, Test_acc:88.5%, Test_loss:0.484
Done

5、结果可视化

import matplotlib.pyplot as plt
#隐藏警告
import warnings
warnings.filterwarnings("ignore")               #忽略警告信息

epochs_range = range(epoches)

plt.figure(figsize=(12, 3))
plt.subplot(1, 2, 1)

plt.plot(epochs_range, train_acc, label='Training Accuracy')
plt.plot(epochs_range, test_acc, label='Test Accuracy')
plt.legend(loc='lower right')
plt.title('Training Accuracy')

plt.subplot(1, 2, 2)
plt.plot(epochs_range, train_loss, label='Training Loss')
plt.plot(epochs_range, test_loss, label='Test Loss')
plt.legend(loc='upper right')
plt.title('Training= Loss')
plt.show()

在这里插入图片描述

探索：

在Resnet模型跑的时候，是跑了80次，但是这个模型并不适合跑30次，在经过20、30、40、80次等实验，与批次大小为16、32、48、64后，发现40次左右是比较较好一点结果；
从测试集来说，这个模型效果高于RenNet模型，从验证集来说，准确率也高于，但是在20轮有一次误差比较大，但是只有一次，属于随机误差。
ResNet实验

6、模型评估

# 将参数加载到model当中 
best_model.load_state_dict(torch.load(PATH, map_location=device)) 
epoch_test_acc, epoch_test_loss = test(test_dl, best_model, loss_fn)

print(epoch_test_acc, epoch_test_loss)