当前位置：首页 > news >正文

深度学习基础--将yolov5的backbone模块用于目标识别会出现怎么效果呢？？

news 2025/9/17 6:28:47

🍨 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

yolov5网络结构比较复杂，上次我们简要介绍了yolov5网络模块，并且复现了C3模块，深度学习基础–yolov5网络结构简介，C3模块构建;
这一次我们将复现backbone模块，将目标检测网络结构用到目标识别上，会是怎样的效果呢？？？；
这周是考试周，周一到周四一直都在准备考试和去考试，昨天开始又发高烧，更新较慢；
欢迎收藏加关注，本人将会持续更新。
文章目录
- 案例
- - 1、数据处理
  - - 1、导入库
    - 2、查看数据类别
    - 3、导入数据
    - 4、数据集划分
    - 5、展示一批数据
  - 2、模型构建
  - 3、模型训练
  - - 1、构建训练集
    - 2、构建测试集
    - 3、设置超参数
  - 4、模型正式训练
  - 5、结果显示和评估
  - - 1、结果显示
    - 2、评估

案例

将backbone模块用于识别天气分类

1、数据处理

1、导入库

import torch 
import torchvision
import torch.nn as nn 
import torchvision.transforms as transforms
from torchvision import datasets, transformsdevice = "cuda" if torch.cuda.is_available() else "cpu"device

'cuda'

2、查看数据类别

import os, pathlib data_dir = './data/'
data_dir = pathlib.Path(data_dir)classnames = [str(path).split("\\")[0] for path in os.listdir(data_dir)]
classnames

['cloudy', 'rain', 'shine', 'sunrise']

3、导入数据

data_transforms = transforms.Compose([transforms.Resize([224, 224]),transforms.ToTensor(),transforms.Normalize(              # 数据标准化处理---> 转化为 标准状态分布，使模型更容易收敛mean=[0.485, 0.456, 0.406],  # rgb，均值std=[0.229, 0.224, 0.225]    # rgb，标准差，这两个从数据集中随机抽样得到的)
])total_data = datasets.ImageFolder("./data/", data_transforms)
total_data

Dataset ImageFolderNumber of datapoints: 1125Root location: ./data/StandardTransform
Transform: Compose(Resize(size=[224, 224], interpolation=bilinear, max_size=None, antialias=True)ToTensor()Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]))

4、数据集划分

训练集：测试集 = 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])
print("train_size", len(train_data))
print("test_size", len(test_data))

train_size 900
test_size 225

# 动态加载数据集
batch_size = 4train_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=True
)

# 查看数据格式
temp_data, temp_label = next(iter(train_dl))print("data: ", temp_data.shape)
print("data_labels: ", temp_label)

data:  torch.Size([4, 3, 224, 224])
data_labels:  tensor([3, 2, 0, 0])

5、展示一批数据

这里一批次大小：4

import matplotlib.pyplot as plt temp_images, temp_labels = next(iter(test_dl))plt.figure(figsize=(20, 10))for i in range(4):plt.subplot(5, 5, i + 1)plt.imshow(temp_images[i].cpu().numpy().transpose(1, 2, 0))  # （C， H， W)  ==>  （H， W， C）plt.title(classnames[temp_labels[i]])plt.axis('off')plt.show()

在这里插入图片描述

2、模型构建

整体网络：

在这里插入图片描述

C3网络参考：深度学习基础–yolov5网络结构简介，C3模块构建

SPPF网络模块图结构如下：

在这里插入图片描述

import torch.nn.functional as F 
import warnings  # 确保导入 warnings 模块# 自动计算p(填充)
def autop(k, p=None):if p is None:p = k // 2 if isinstance(k, int) else [i // 2 for i in k]return p # Conv模块搭建
'''
卷积层 + 标准化 + 激活函数
'''
class Conv(nn.Module):def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):super().__init__()'''groups: 1: 标准卷积c1:  深度卷积1 ~ c1: 分组卷积bias:false: 不使用偏置'''self.conv = nn.Conv2d(c1, c2, kernel_size=k, stride=s, padding=autop(k, p), groups=g, bias=False)  self.bn = nn.BatchNorm2d(c2)'''act:true: silu激活函数否则: 如果是nn.Mudule(如: nn.Relu), 则调用本身否则: Identity, 什么都不做'''self.act = nn.SiLU() if act is True else (act if isinstance(act, nn.Module()) else nn.Identity())def forward(self,x):return self.act(self.bn(self.conv(x)))# Bottleneck模块, 用于特征提取和用于防止梯度消失、梯度爆炸问题
class Bottleneck(nn.Module):def __init__(self, c1, c2, shortcut=True, g=1, e=0.5):    # shortcut: 是否需要残差连接， e: 模型深度super().__init__()c_ = int(c1 * 2)self.cv1 = Conv(c1, c_, 1, 1)self.cv2 = Conv(c_, c2, 3, 1, g=g)self.add = shortcut and c1 == c2  def forward(self, x):return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))# 搭建C3模块
class C3(nn.Module):def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):super().__init__()'''刚开始: 卷积层2层后面: n层bottlenck后 concat后 conv'''c_ = int(c1 * e)self.cv1 = Conv(c1, c_, 1, 1)self.cv2 = Conv(c1, c_, 1, 1)   # 用于拼接self.cv3 = Conv(2 * c_, c2, 1)self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))  # * 解包def forward(self, x):# 拼接，按照 dim=1维进行拼接，故列要相同return self.cv3(torch.cat([self.m(self.cv1(x)), self.cv2(x)], dim=1))   # 结合图就知道了结构# 搭建SPPF模块，用于特征融合
class SPPF(nn.Module):def __init__(self, c1, c2, k=5):super().__init__()c_ = c1 // 2 self.cv1 = Conv(c1, c_, 1, 1)self.cv2 = Conv(c_ * 4, c2, 1, 1)   # 模型融合，这个时候模型通道扩大4倍，套用池化层公式，发现通过.m 通道数数不变self.m = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)   # 套用卷积层、池化层公式，发现输出通道不变def forward(self, x):x = self.cv1(x)with warnings.catch_warnings():warnings.simplefilter('ignore')y1 = self.m(x)y2 = self.m(y1)y3 = self.m(y2)return self.cv2(torch.cat([x, y1, y2, y3], 1))   # 结合图# 搭建backbone模块
class Yolov5_backbone(nn.Module):def __init__(self):super(Yolov5_backbone, self).__init__()# 采用常规卷积, kernel_size, stride 与 yolov5.yaml配置文件一致self.conv_1 = Conv(3, 64, 3, 2, 2) self.conv_2 = Conv(64, 128, 3, 2)self.c3_3 = C3(128, 128)self.conv_4 = Conv(128, 256, 3, 2)self.c3_5 = C3(256, 256)self.conv_6 = Conv(256, 512, 3, 2)self.c3_7 = C3(512, 512)self.conv_8 = Conv(512, 1024, 3, 2)self.c3_9 = C3(1024, 1024)self.SPPF_10 = SPPF(1024, 1024, 5)self.classifiler = nn.Sequential(nn.Linear(in_features=65536, out_features=100),nn.ReLU(),nn.Linear(in_features=100, out_features=len(classnames)))def forward(self, x):x = self.conv_1(x)x = self.conv_2(x)x = self.c3_3(x)x = self.conv_4(x)x = self.c3_5(x)x = self.conv_6(x)x = self.c3_7(x)x = self.conv_8(x)x = self.c3_9(x)x = self.SPPF_10(x)x = torch.flatten(x, start_dim=1)x = self.classifiler(x)return x

# 输出参数
model = Yolov5_backbone().to(device)
model

Yolov5_backbone((conv_1): Conv((conv): Conv2d(3, 64, kernel_size=(3, 3), stride=(2, 2), padding=(2, 2), bias=False)(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(conv_2): Conv((conv): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(c3_3): C3((cv1): Conv((conv): Conv2d(128, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv2): Conv((conv): Conv2d(128, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv3): Conv((conv): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(m): Sequential((0): Bottleneck((cv1): Conv((conv): Conv2d(64, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv2): Conv((conv): Conv2d(128, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU()))))(conv_4): Conv((conv): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(c3_5): C3((cv1): Conv((conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv2): Conv((conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv3): Conv((conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(m): Sequential((0): Bottleneck((cv1): Conv((conv): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv2): Conv((conv): Conv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU()))))(conv_6): Conv((conv): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(c3_7): C3((cv1): Conv((conv): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv2): Conv((conv): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv3): Conv((conv): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(m): Sequential((0): Bottleneck((cv1): Conv((conv): Conv2d(256, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv2): Conv((conv): Conv2d(512, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU()))))(conv_8): Conv((conv): Conv2d(512, 1024, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)(bn): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(c3_9): C3((cv1): Conv((conv): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv2): Conv((conv): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv3): Conv((conv): Conv2d(1024, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(m): Sequential((0): Bottleneck((cv1): Conv((conv): Conv2d(512, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv2): Conv((conv): Conv2d(1024, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU()))))(SPPF_10): SPPF((cv1): Conv((conv): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(cv2): Conv((conv): Conv2d(2048, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)(bn): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(act): SiLU())(m): MaxPool2d(kernel_size=5, stride=1, padding=2, dilation=1, ceil_mode=False))(classifiler): Sequential((0): Linear(in_features=65536, out_features=100, bias=True)(1): ReLU()(2): Linear(in_features=100, out_features=4, bias=True))
)

import torchsummary as summary 
summary.summary(model, (3, 224, 224))

----------------------------------------------------------------Layer (type)               Output Shape         Param #
================================================================Conv2d-1         [-1, 64, 113, 113]           1,728BatchNorm2d-2         [-1, 64, 113, 113]             128SiLU-3         [-1, 64, 113, 113]               0Conv-4         [-1, 64, 113, 113]               0Conv2d-5          [-1, 128, 57, 57]          73,728BatchNorm2d-6          [-1, 128, 57, 57]             256SiLU-7          [-1, 128, 57, 57]               0Conv-8          [-1, 128, 57, 57]               0Conv2d-9           [-1, 64, 57, 57]           8,192BatchNorm2d-10           [-1, 64, 57, 57]             128SiLU-11           [-1, 64, 57, 57]               0Conv-12           [-1, 64, 57, 57]               0Conv2d-13          [-1, 128, 57, 57]           8,192BatchNorm2d-14          [-1, 128, 57, 57]             256SiLU-15          [-1, 128, 57, 57]               0Conv-16          [-1, 128, 57, 57]               0Conv2d-17           [-1, 64, 57, 57]          73,728BatchNorm2d-18           [-1, 64, 57, 57]             128SiLU-19           [-1, 64, 57, 57]               0Conv-20           [-1, 64, 57, 57]               0Bottleneck-21           [-1, 64, 57, 57]               0Conv2d-22           [-1, 64, 57, 57]           8,192BatchNorm2d-23           [-1, 64, 57, 57]             128SiLU-24           [-1, 64, 57, 57]               0Conv-25           [-1, 64, 57, 57]               0Conv2d-26          [-1, 128, 57, 57]          16,384BatchNorm2d-27          [-1, 128, 57, 57]             256SiLU-28          [-1, 128, 57, 57]               0Conv-29          [-1, 128, 57, 57]               0C3-30          [-1, 128, 57, 57]               0Conv2d-31          [-1, 256, 29, 29]         294,912BatchNorm2d-32          [-1, 256, 29, 29]             512SiLU-33          [-1, 256, 29, 29]               0Conv-34          [-1, 256, 29, 29]               0Conv2d-35          [-1, 128, 29, 29]          32,768BatchNorm2d-36          [-1, 128, 29, 29]             256SiLU-37          [-1, 128, 29, 29]               0Conv-38          [-1, 128, 29, 29]               0Conv2d-39          [-1, 256, 29, 29]          32,768BatchNorm2d-40          [-1, 256, 29, 29]             512SiLU-41          [-1, 256, 29, 29]               0Conv-42          [-1, 256, 29, 29]               0Conv2d-43          [-1, 128, 29, 29]         294,912BatchNorm2d-44          [-1, 128, 29, 29]             256SiLU-45          [-1, 128, 29, 29]               0Conv-46          [-1, 128, 29, 29]               0Bottleneck-47          [-1, 128, 29, 29]               0Conv2d-48          [-1, 128, 29, 29]          32,768BatchNorm2d-49          [-1, 128, 29, 29]             256SiLU-50          [-1, 128, 29, 29]               0Conv-51          [-1, 128, 29, 29]               0Conv2d-52          [-1, 256, 29, 29]          65,536BatchNorm2d-53          [-1, 256, 29, 29]             512SiLU-54          [-1, 256, 29, 29]               0Conv-55          [-1, 256, 29, 29]               0C3-56          [-1, 256, 29, 29]               0Conv2d-57          [-1, 512, 15, 15]       1,179,648BatchNorm2d-58          [-1, 512, 15, 15]           1,024SiLU-59          [-1, 512, 15, 15]               0Conv-60          [-1, 512, 15, 15]               0Conv2d-61          [-1, 256, 15, 15]         131,072BatchNorm2d-62          [-1, 256, 15, 15]             512SiLU-63          [-1, 256, 15, 15]               0Conv-64          [-1, 256, 15, 15]               0Conv2d-65          [-1, 512, 15, 15]         131,072BatchNorm2d-66          [-1, 512, 15, 15]           1,024SiLU-67          [-1, 512, 15, 15]               0Conv-68          [-1, 512, 15, 15]               0Conv2d-69          [-1, 256, 15, 15]       1,179,648BatchNorm2d-70          [-1, 256, 15, 15]             512SiLU-71          [-1, 256, 15, 15]               0Conv-72          [-1, 256, 15, 15]               0Bottleneck-73          [-1, 256, 15, 15]               0Conv2d-74          [-1, 256, 15, 15]         131,072BatchNorm2d-75          [-1, 256, 15, 15]             512SiLU-76          [-1, 256, 15, 15]               0Conv-77          [-1, 256, 15, 15]               0Conv2d-78          [-1, 512, 15, 15]         262,144BatchNorm2d-79          [-1, 512, 15, 15]           1,024SiLU-80          [-1, 512, 15, 15]               0Conv-81          [-1, 512, 15, 15]               0C3-82          [-1, 512, 15, 15]               0Conv2d-83           [-1, 1024, 8, 8]       4,718,592BatchNorm2d-84           [-1, 1024, 8, 8]           2,048SiLU-85           [-1, 1024, 8, 8]               0Conv-86           [-1, 1024, 8, 8]               0Conv2d-87            [-1, 512, 8, 8]         524,288BatchNorm2d-88            [-1, 512, 8, 8]           1,024SiLU-89            [-1, 512, 8, 8]               0Conv-90            [-1, 512, 8, 8]               0Conv2d-91           [-1, 1024, 8, 8]         524,288BatchNorm2d-92           [-1, 1024, 8, 8]           2,048SiLU-93           [-1, 1024, 8, 8]               0Conv-94           [-1, 1024, 8, 8]               0Conv2d-95            [-1, 512, 8, 8]       4,718,592BatchNorm2d-96            [-1, 512, 8, 8]           1,024SiLU-97            [-1, 512, 8, 8]               0Conv-98            [-1, 512, 8, 8]               0Bottleneck-99            [-1, 512, 8, 8]               0Conv2d-100            [-1, 512, 8, 8]         524,288BatchNorm2d-101            [-1, 512, 8, 8]           1,024SiLU-102            [-1, 512, 8, 8]               0Conv-103            [-1, 512, 8, 8]               0Conv2d-104           [-1, 1024, 8, 8]       1,048,576BatchNorm2d-105           [-1, 1024, 8, 8]           2,048SiLU-106           [-1, 1024, 8, 8]               0Conv-107           [-1, 1024, 8, 8]               0C3-108           [-1, 1024, 8, 8]               0Conv2d-109            [-1, 512, 8, 8]         524,288BatchNorm2d-110            [-1, 512, 8, 8]           1,024SiLU-111            [-1, 512, 8, 8]               0Conv-112            [-1, 512, 8, 8]               0MaxPool2d-113            [-1, 512, 8, 8]               0MaxPool2d-114            [-1, 512, 8, 8]               0MaxPool2d-115            [-1, 512, 8, 8]               0Conv2d-116           [-1, 1024, 8, 8]       2,097,152BatchNorm2d-117           [-1, 1024, 8, 8]           2,048SiLU-118           [-1, 1024, 8, 8]               0Conv-119           [-1, 1024, 8, 8]               0SPPF-120           [-1, 1024, 8, 8]               0Linear-121                  [-1, 100]       6,553,700ReLU-122                  [-1, 100]               0Linear-123                    [-1, 4]             404
================================================================
Total params: 25,213,112
Trainable params: 25,213,112
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 149.98
Params size (MB): 96.18
Estimated Total Size (MB): 246.74
----------------------------------------------------------------

3、模型训练

1、构建训练集

def train(dataloader, model, loss_fn, optimizer):size = len(dataloader.dataset)  # 总数目num_batch = len(dataloader)      # 批次数目train_acc, train_loss = 0, 0for X, y in dataloader:X, y = X.to(device), y.to(device) predict = model(X)loss = loss_fn(predict, y)# 梯度清0、求导、重新设置参数optimizer.zero_grad()loss.backward()optimizer.step()train_acc += (predict.argmax(1) == y).type(torch.float).sum().item()train_loss += loss.item()train_acc /= sizetrain_loss /= num_batchreturn train_acc, train_loss

2、构建测试集

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

3、设置超参数

learn_rate = 1e-4 
optimizer = torch.optim.Adam(model.parameters(), lr=learn_rate)
loss_fn = nn.CrossEntropyLoss()

4、模型正式训练

import copy epochs = 60train_acc, train_loss, test_acc, test_loss = [], [], [], []best_acc = 0for epoch in range(epochs):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_modelif epoch_test_acc > best_acc:best_acc   = epoch_test_accbest_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']template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}, Lr:{:.2E}')print(template.format(epoch+1, epoch_train_acc*100, epoch_train_loss, epoch_test_acc*100, epoch_test_loss, lr))# 保存最佳模型到文件中
PATH = './best_model.pth'  # 保存的参数文件名
torch.save(model.state_dict(), PATH)print('Done')

Epoch: 1, Train_acc:59.0%, Train_loss:1.106, Test_acc:71.1%, Test_loss:0.740, Lr:1.00E-04
Epoch: 2, Train_acc:70.0%, Train_loss:0.776, Test_acc:83.1%, Test_loss:0.468, Lr:1.00E-04
Epoch: 3, Train_acc:75.2%, Train_loss:0.661, Test_acc:84.0%, Test_loss:0.461, Lr:1.00E-04
Epoch: 4, Train_acc:77.4%, Train_loss:0.605, Test_acc:88.4%, Test_loss:0.396, Lr:1.00E-04
Epoch: 5, Train_acc:80.3%, Train_loss:0.529, Test_acc:82.7%, Test_loss:0.415, Lr:1.00E-04
Epoch: 6, Train_acc:83.8%, Train_loss:0.422, Test_acc:83.6%, Test_loss:0.416, Lr:1.00E-04
Epoch: 7, Train_acc:85.8%, Train_loss:0.423, Test_acc:87.1%, Test_loss:0.343, Lr:1.00E-04
Epoch: 8, Train_acc:85.6%, Train_loss:0.393, Test_acc:87.6%, Test_loss:0.306, Lr:1.00E-04
Epoch: 9, Train_acc:86.3%, Train_loss:0.354, Test_acc:89.3%, Test_loss:0.338, Lr:1.00E-04
Epoch:10, Train_acc:86.6%, Train_loss:0.340, Test_acc:92.9%, Test_loss:0.276, Lr:1.00E-04
Epoch:11, Train_acc:88.8%, Train_loss:0.317, Test_acc:90.2%, Test_loss:0.290, Lr:1.00E-04
Epoch:12, Train_acc:87.9%, Train_loss:0.327, Test_acc:88.0%, Test_loss:0.338, Lr:1.00E-04
Epoch:13, Train_acc:89.2%, Train_loss:0.315, Test_acc:92.0%, Test_loss:0.337, Lr:1.00E-04
Epoch:14, Train_acc:91.1%, Train_loss:0.230, Test_acc:92.0%, Test_loss:0.369, Lr:1.00E-04
Epoch:15, Train_acc:93.3%, Train_loss:0.182, Test_acc:89.8%, Test_loss:0.278, Lr:1.00E-04
Epoch:16, Train_acc:91.6%, Train_loss:0.229, Test_acc:90.2%, Test_loss:0.290, Lr:1.00E-04
Epoch:17, Train_acc:90.9%, Train_loss:0.230, Test_acc:91.6%, Test_loss:0.272, Lr:1.00E-04
Epoch:18, Train_acc:93.9%, Train_loss:0.152, Test_acc:92.0%, Test_loss:0.280, Lr:1.00E-04
Epoch:19, Train_acc:94.7%, Train_loss:0.159, Test_acc:92.0%, Test_loss:0.262, Lr:1.00E-04
Epoch:20, Train_acc:95.9%, Train_loss:0.124, Test_acc:91.1%, Test_loss:0.260, Lr:1.00E-04
Epoch:21, Train_acc:95.7%, Train_loss:0.102, Test_acc:88.9%, Test_loss:0.342, Lr:1.00E-04
Epoch:22, Train_acc:95.9%, Train_loss:0.113, Test_acc:92.4%, Test_loss:0.275, Lr:1.00E-04
Epoch:23, Train_acc:96.1%, Train_loss:0.130, Test_acc:92.9%, Test_loss:0.308, Lr:1.00E-04
Epoch:24, Train_acc:94.8%, Train_loss:0.161, Test_acc:86.7%, Test_loss:0.456, Lr:1.00E-04
Epoch:25, Train_acc:95.2%, Train_loss:0.139, Test_acc:89.3%, Test_loss:0.428, Lr:1.00E-04
Epoch:26, Train_acc:96.0%, Train_loss:0.103, Test_acc:92.9%, Test_loss:0.313, Lr:1.00E-04
Epoch:27, Train_acc:96.0%, Train_loss:0.098, Test_acc:88.9%, Test_loss:0.520, Lr:1.00E-04
Epoch:28, Train_acc:97.2%, Train_loss:0.079, Test_acc:91.1%, Test_loss:0.404, Lr:1.00E-04
Epoch:29, Train_acc:98.6%, Train_loss:0.037, Test_acc:92.0%, Test_loss:0.270, Lr:1.00E-04
Epoch:30, Train_acc:98.8%, Train_loss:0.033, Test_acc:88.4%, Test_loss:0.520, Lr:1.00E-04
Epoch:31, Train_acc:95.6%, Train_loss:0.139, Test_acc:91.6%, Test_loss:0.370, Lr:1.00E-04
Epoch:32, Train_acc:96.7%, Train_loss:0.116, Test_acc:89.3%, Test_loss:0.376, Lr:1.00E-04
Epoch:33, Train_acc:96.4%, Train_loss:0.102, Test_acc:91.6%, Test_loss:0.342, Lr:1.00E-04
Epoch:34, Train_acc:98.6%, Train_loss:0.049, Test_acc:87.1%, Test_loss:0.417, Lr:1.00E-04
Epoch:35, Train_acc:97.9%, Train_loss:0.068, Test_acc:90.7%, Test_loss:0.423, Lr:1.00E-04
Epoch:36, Train_acc:98.3%, Train_loss:0.048, Test_acc:89.3%, Test_loss:0.492, Lr:1.00E-04
Epoch:37, Train_acc:98.0%, Train_loss:0.054, Test_acc:91.1%, Test_loss:0.355, Lr:1.00E-04
Epoch:38, Train_acc:98.6%, Train_loss:0.060, Test_acc:92.4%, Test_loss:0.402, Lr:1.00E-04
Epoch:39, Train_acc:97.9%, Train_loss:0.065, Test_acc:86.7%, Test_loss:0.498, Lr:1.00E-04
Epoch:40, Train_acc:98.0%, Train_loss:0.055, Test_acc:88.4%, Test_loss:0.514, Lr:1.00E-04
Epoch:41, Train_acc:99.1%, Train_loss:0.029, Test_acc:90.7%, Test_loss:0.381, Lr:1.00E-04
Epoch:42, Train_acc:98.0%, Train_loss:0.069, Test_acc:92.4%, Test_loss:0.377, Lr:1.00E-04
Epoch:43, Train_acc:99.4%, Train_loss:0.021, Test_acc:90.2%, Test_loss:0.403, Lr:1.00E-04
Epoch:44, Train_acc:98.0%, Train_loss:0.055, Test_acc:85.3%, Test_loss:0.686, Lr:1.00E-04
Epoch:45, Train_acc:98.0%, Train_loss:0.074, Test_acc:91.1%, Test_loss:0.321, Lr:1.00E-04
Epoch:46, Train_acc:98.6%, Train_loss:0.038, Test_acc:91.6%, Test_loss:0.426, Lr:1.00E-04
Epoch:47, Train_acc:97.4%, Train_loss:0.075, Test_acc:87.1%, Test_loss:0.604, Lr:1.00E-04
Epoch:48, Train_acc:99.6%, Train_loss:0.027, Test_acc:91.6%, Test_loss:0.379, Lr:1.00E-04
Epoch:49, Train_acc:99.8%, Train_loss:0.007, Test_acc:92.4%, Test_loss:0.381, Lr:1.00E-04
Epoch:50, Train_acc:100.0%, Train_loss:0.007, Test_acc:92.9%, Test_loss:0.361, Lr:1.00E-04
Epoch:51, Train_acc:99.8%, Train_loss:0.018, Test_acc:90.7%, Test_loss:0.446, Lr:1.00E-04
Epoch:52, Train_acc:99.0%, Train_loss:0.032, Test_acc:89.8%, Test_loss:0.588, Lr:1.00E-04
Epoch:53, Train_acc:97.7%, Train_loss:0.060, Test_acc:90.7%, Test_loss:0.456, Lr:1.00E-04
Epoch:54, Train_acc:97.7%, Train_loss:0.059, Test_acc:89.8%, Test_loss:0.506, Lr:1.00E-04
Epoch:55, Train_acc:98.6%, Train_loss:0.046, Test_acc:90.7%, Test_loss:0.350, Lr:1.00E-04
Epoch:56, Train_acc:99.7%, Train_loss:0.010, Test_acc:91.6%, Test_loss:0.349, Lr:1.00E-04
Epoch:57, Train_acc:99.6%, Train_loss:0.012, Test_acc:91.6%, Test_loss:0.369, Lr:1.00E-04
Epoch:58, Train_acc:98.9%, Train_loss:0.053, Test_acc:88.9%, Test_loss:0.666, Lr:1.00E-04
Epoch:59, Train_acc:98.2%, Train_loss:0.054, Test_acc:87.1%, Test_loss:0.509, Lr:1.00E-04
Epoch:60, Train_acc:98.7%, Train_loss:0.037, Test_acc:90.2%, Test_loss:0.513, Lr:1.00E-04
Done

5、结果显示和评估

1、结果显示

import matplotlib.pyplot as plt #隐藏警告和显示中文
import warnings
warnings.filterwarnings("ignore")               #忽略警告信息
plt.rcParams['font.sans-serif']    = ['SimHei'] # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False      # 用来正常显示负号
plt.rcParams['figure.dpi']         = 100        #分辨率x = range(epochs)
# 创建画板
plt.figure(figsize=(12, 3))
# 子图一
plt.subplot(1, 2, 1)
plt.plot(x, train_acc, label='Train Accurary')
plt.plot(x, test_acc, label='Test Accurary')
plt.legend(loc='lower right')
plt.title("Train and test Accurary")
# 子图二
plt.subplot(1, 2, 2)
plt.plot(x, train_loss, label='Train loss')
plt.plot(x, test_loss, label='Test loss')
plt.legend(loc='upper right')
plt.title("Train and test Loss")plt.show()

在这里插入图片描述

👀 解释：

总体效果还是不错的，损失率低于1，但是测试集的损失率有点小小不稳定；
准确率：刚开始出现了欠拟合的现象，但是后面好了，训练准确率稳定在100%附件(98%、99%等)，测试集稳定在90%附件；
整体：yolov5这个用于目标检测的网络用语目标识别也是有不错的效果。

2、评估

best_model.load_state_dict(torch.load(PATH, map_location=device))
epoch_test_acc, epoch_test_loss = test(test_dl, best_model, loss_fn)epoch_test_acc, epoch_test_loss

(0.9022222222222223, 0.5129974257313852)

准确率在0.9左右，效果良好，且损失率为0.5，低于1.0。

深度学习基础--将yolov5的backbone模块用于目标识别会出现怎么效果呢？？

🍨 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 yolov5网络结构比较复杂，上次我们简要介绍了yolov5网络模块，并且复现了C3模块，深度学习基础–yolov5网络结构简介&a…...

编程日记 2024/12/19 14:29:55

操作系统（16）I/O软件

前言操作系统I/O软件是负责管理和控制计算机系统与外围设备（例如键盘、鼠标、打印机、存储设备等）之间交互的软件。一、I/O软件的定义与功能定义：I/O软件，也称为输入/输出软件，是计算机系统中用于管理和控制设备与主…...

编程日记 2024/12/19 14:27:52

leetcode437.路径总和III

标签：前缀和问题：给定一个二叉树的根节点 root ，和一个整数 targetSum ，求该二叉树里节点值之和等于 targetSum 的路径的数目。路径不需要从根节点开始，也不需要在叶子节点结束，但是路径方向必须是向下…...

编程日记 2024/12/19 14:24:50

WebGPU、WebGL 和 OpenGL/Vulkan对比分析

WebGPU、WebGL 和 OpenGL/Vulkan 都是用于图形渲染和计算的图形API，但它们的设计理念、功能和适用场景有所不同。以下是它们的总结和对比分析： 1. WebGPU WebGPU 是一个新的、现代化的图形和计算API，设计目的是为Web平台提供更接近硬件的性…...

编程日记 2024/12/19 14:22:48

不可重入锁与死锁

不可重入锁确实可能导致死锁，特别是在同一线程尝试多次获取同一把锁时。如果锁是不可重入的，那么线程在第二次尝试获取锁时会永远阻塞，从而导致死锁。不可重入锁与死锁的关系不可重入锁不允许同一个线程多次获取同一把锁。在以下情况下&am…...

编程日记 2024/12/19 14:21:47

XXE-Lab靶场漏洞复现

1.尝试登录输入账号admin/密码admin进行登录，并未有页面进行跳转 2.尝试抓包分析请求包数据我们可以发现页面中存在xml请求，我们就可以构造我们的xml请求语句来获取想要的数据 3.构造语句 <?xml version"1.0" ?> <!DOCTYPE fo…...

编程日记 2024/12/19 14:19:45

从Windows到Linux：跨平台数据库备份与还原

数据库的备份与还原目录引言备份 2.1 备份所有数据库2.2 备份单个数据库2.3 备份多个指定数据库传输备份文件还原 4.1 还原所有数据库4.2 还原单个数据库4.3 还原多个指定数据库注意事项拓展 1. 引言在不同的操作系统间进行数据库迁移时，命令行工具是我们的…...

编程日记 2024/12/19 14:17:42

upload-labs

Win平台靶场靶场2 教程教程教程 pass-01 bash 本pass在客户端使用js对不合法图片进行检查！前端绕过, 禁用前端js代码, 或者上传图片, 抓包改后缀为 php , 后端没有校验 bash POST /Pass-01/index.php HTTP/1.1 Host: 47.122.3.214:8889 Content-Length: 49…...

编程日记 2024/12/19 14:14:37

【西门子PLC.博途】——面向对象编程及输入输出映射FC块

当我们做面向对象编程的时候，需要用到输入输出的映射。这样建立的变量就能够被复用，从而最大化利用了我们建立的udt对象。下面就来讲讲映射是什么。从本质上来说，映射就是拿实际物理对象对应程序虚拟对象，假设程序对象是I0.0&…...

编程日记 2024/12/19 14:12:30

牛客周赛 Round 72 题解

本次牛客最后一个线段树之前我也没碰到过，等后续复习到线段树再把那个题当例题发出来小红的01串（一） 思路：正常模拟，从前往后遍历一遍去统计即可 #include<bits/stdc.h> using namespace std; #define int lo…...

编程日记 2024/12/19 14:07:24

Flux Tools 结构简析

Flux Tools 结构简析 BFL 这次一共发布了 Canny、Depth、Redux、Fill 四个 Tools 模型系列，分别对应我们熟悉的 ControlNets、Image Variation（IP Adapter）和 Inpainting 三种图片条件控制方法。虽然实现功能是相同的，但是其具体…...

编程日记 2024/12/19 14:05:23

0 前言

ArCS作为一个基于Rust的CAD（计算机辅助设计）开源系统，尽管已经有四年未更新，但其设计理念和技术实现仍然具有很高的学习和参考价值。以下是对ArCS项目的进一步分析和解读： 一、项目亮点与技术优势高效与安全的Rust语…...

编程日记 2024/12/19 14:02:19

ARM嵌入式学习--第八天（PWM）

PWM -PWM介绍 PWM（pulse Width Modulation）简称脉宽调制，是利用微处理器的数字输出来对模拟电路进行控制的一种非常有效的技术，广泛应用在测量，通信，工控等方面 PWM的频率是指在1秒钟内，信号从…...

编程日记 2024/12/19 14:00:15

遇到“REMOTE HOST IDENTIFICATION HAS CHANGED!”（远程主机识别已更改）的警告

连接虚拟机时提示报错： [insocoperhq-soc-cap-raw3 ~]$ ssh root10.99.141.104WARNING: REMOTE HOST IDENTIFICATION HAS CHANGED! IT IS POSSIBLE THAT SOMEONE IS DOING SOMETHING NASTY! Someone could be eavesdropping on you right now (man-in-the-midd…...

编程日记 2024/12/19 13:56:07

vue3前端组件库的搭建与发布（一）

前言： 最近在做公司项目中，有这么一件事情，很是头疼，就是同一套代码，不同项目，要改相同bug，改好多遍，改的都想吐，于是就想做一个组件库，这样更新一下就全都可…...

编程日记 2024/12/19 13:51:00

COMSOL快捷键及内置函数

文章目录 COMSOL快捷键使用COMSOL算子求最大值和最小值COMSOL内置函数3.1 解析函数3.2 插值函数3.3 分段函数3.4 高斯脉冲函数3.5 斜坡函数3.6 矩形函数3.7 波形函数3.8 随机函数3.9 Matlab函数3.10 SWITCH函数 COMSOL快捷键 Ctrl＋/ 可快速打开预定义的物理量列表。…...

编程日记 2024/12/19 13:48:58

HUAWEI-eNSP交换机链路聚合(手动负载分担模式)

配置思路：HUAWEI交换机链路聚合有LACP模式跟手动负载分担模式，本文主打手动负载分担模式：首先交换机-PC之间划分基本vlan，交换机-交换机之间创建链路聚合组，划分端口至链路聚合分组（缺省模式为手动负载分担模式）。结果验证要求同vlan可以ping通，关闭某个聚合端口后仍可…...

编程日记 2024/12/19 13:46:51

番外篇 | Hyper-YOLO：超图计算与YOLO架构相结合成为目标检测新的SOTA ！

前言：Hello大家好，我是小哥谈。Hyper-YOLO，该方法融合了超图计算以捕捉视觉特征之间复杂的高阶关联。传统的YOLO模型虽然功能强大，但其颈部设计存在局限性，限制了跨层特征的融合以及高阶特征关系的利用。Hyper-YOLO在骨干和颈部的联合增强下，成为一个突破性的架构。在COC…...

编程日记 2024/12/19 13:45:50

【MATLAB第109期】基于MATLAB的带置信区间的RSA区域敏感性分析方法，无目标函数

【MATLAB第108期】基于MATLAB的带置信区间的RSA区域敏感性分析方法，无目标函数参考第64期文章【MATLAB第64期】【保姆级教程】基于MATLAB的SOBOL全局敏感性分析模型运用（含无目标函数，考虑代理模型） 创新点： 1、采…...

编程日记 2024/12/19 13:39:44

Bootstrap 表格

Bootstrap 表格引言 Bootstrap 是一个流行的前端框架，它提供了一套丰富的工具和组件，用于快速开发响应式和移动设备优先的网页。在本文中，我们将重点讨论 Bootstrap 中的表格组件，包括其基本结构、样式以及如何使用 Bootstrap …...

编程日记 2024/12/19 13:33:36

智慧医疗能源事业线深度画像分析（上）

引言医疗行业作为现代社会的关键基础设施，其能源消耗与环境影响正日益受到关注。随着全球"双碳"目标的推进和可持续发展理念的深入，智慧医疗能源事业线应运而生，致力于通过创新技术与管理方案，重构医疗领域的能源使用模式。这一事业线融合了能源管理、可持续发…...

编程新知 2025/9/16 1:45:47

AI书签管理工具开发全记录（十九）：嵌入资源处理

1.前言 📝 在上一篇文章中，我们完成了书签的导入导出功能。本篇文章我们研究如何处理嵌入资源，方便后续将资源打包到一个可执行文件中。 2.embed介绍 🎯 Go 1.16 引入了革命性的 embed 包，彻底改变了静态资源管理的…...

编程新知 2025/9/9 18:41:56

初探Service服务发现机制

1.Service简介 Service是将运行在一组Pod上的应用程序发布为网络服务的抽象方法。主要功能：服务发现和负载均衡。 Service类型的包括ClusterIP类型、NodePort类型、LoadBalancer类型、ExternalName类型 2.Endpoints简介 Endpoints是一种Kubernetes资源&#xf…...

编程新知 2025/9/2 19:56:38

Qemu arm操作系统开发环境

使用qemu虚拟arm硬件比较合适。步骤如下： 安装qemu apt install qemu-system安装aarch64-none-elf-gcc 需要手动下载，下载地址：https://developer.arm.com/-/media/Files/downloads/gnu/13.2.rel1/binrel/arm-gnu-toolchain-13.2.rel1-x…...

编程新知 2025/9/4 22:48:54

论文阅读笔记——Muffin: Testing Deep Learning Libraries via Neural Architecture Fuzzing

Muffin 论文现有方法 CRADLE 和 LEMON，依赖模型推理阶段输出进行差分测试，但在训练阶段是不可行的，因为训练阶段直到最后才有固定输出，中间过程是不断变化的。API 库覆盖低，因为各个 API 都是在各种具体场景下使用。…...

编程新知 2025/9/7 8:50:08

【C++】纯虚函数类外可以写实现吗？

1. 答案先说答案，可以。 2.代码测试 .h头文件 #include <iostream> #include <string>// 抽象基类 class AbstractBase { public:AbstractBase() default;virtual ~AbstractBase() default; // 默认析构函数public:virtual int PureVirtualFunct…...

编程新知 2025/9/9 10:08:11

基于鸿蒙(HarmonyOS5)的打车小程序

1. 开发环境准备安装DevEco Studio (鸿蒙官方IDE)配置HarmonyOS SDK申请开发者账号和必要的API密钥 2. 项目结构设计 ├── entry │ ├── src │ │ ├── main │ │ │ ├── ets │ │ │ │ ├── pages │ │ │ │ │ ├── H…...

编程新知 2025/7/7 16:04:22