第P9周：YOLOv5-Backbone模块实现

最新推荐文章于 2024-10-18 21:44:50 发布

帅先森！

最新推荐文章于 2024-10-18 21:44:50 发布

阅读量381

点赞数 4

分类专栏： pytorch入门实战文章标签： pytorch 人工智能 python YOLO

本文链接：https://blog.csdn.net/weixin_52252887/article/details/140726500

版权

pytorch入门实战专栏收录该内容

9 篇文章 0 订阅

订阅专栏

>- **🍨 本文为[🔗365天深度学习训练营](https://mp.weixin.qq.com/s/0dvHCaOoFnW8SCp3JpzKxg) 中的学习记录博客**
>- **🍖 原作者：[K同学啊](https://mtyjkh.blog.csdn.net/)**

语言环境：Python3.8
编译器：Jupyter Lab
深度学习环境：
- torch==2.3.1+cpu
- torchvision==0.18.1+cpu

4.1 Loss 与 Accuracy 图

4.2 模型评估

1. 前期准备

1. 1设置cpu：

import torch
import torch.nn as nn
import matplotlib.pyplot as plt
import torchvision

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

device

代码输出：

device(type='cpu')

1.2 导入数据

import os,PIL,random,pathlib

data_dir = './weather_photos/'
data_dir = pathlib.Path(data_dir)

data_paths  = list(data_dir.glob('*'))
classeNames = [str(path).split("\\")[1] for path in data_paths]
classeNames

代码输出：

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

# 关于transforms.Compose的更多介绍可以参考：https://blog.csdn.net/qq_38251616/article/details/124878863
train_transforms = transforms.Compose([
    transforms.Resize([224, 224]),  # 将输入图片resize成统一尺寸
    transforms.RandomHorizontalFlip(), # 随机水平翻转
    transforms.ToTensor(),          # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
    transforms.Normalize(           # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
        mean=[0.485, 0.456, 0.406], 
        std=[0.229, 0.224, 0.225])  # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])

total_data = datasets.ImageFolder("./weather_photos/",transform=train_transforms)
total_data

1.3 划分数据集

# train_size表示训练集大小，通过将总体数据长度的80%转换为整数得到；
# test_size表示测试集大小，是总体数据长度减去训练集大小。
train_size = int(0.8 * len(total_data))
test_size  = len(total_data) - train_size

# 使用torch.utils.data.random_split()方法进行数据集划分。该方法将总体数据total_data按照指定的大小比例（[train_size, test_size]）随机划分为训练集和测试集，
# 并将划分结果分别赋值给train_dataset和test_dataset两个变量。
train_dataset, test_dataset = torch.utils.data.random_split(total_data, [train_size, test_size])

batch_size = 4

train_dl = torch.utils.data.DataLoader(train_dataset,
                                       batch_size=batch_size,
                                       shuffle=True,
                                       num_workers=1)
test_dl = torch.utils.data.DataLoader(test_dataset,
                                      batch_size=batch_size,
                                      shuffle=True,
                                      num_workers=1)

for X, y in test_dl:
    print("Shape of X [N, C, H, W]: ", X.shape)
    print("Shape of y: ", y.shape, y.dtype)
    break

2. 搭建包含Backbone模块的模型

2.1 搭建模型

import torch.nn.functional as F

def autopad(k, p=None):  # kernel, padding
    # Pad to 'same'
    if p is None:
        p = k // 2 if isinstance(k, int) else [x // 2 for x in k]  # auto-pad
    return p

class Conv(nn.Module):
    # Standard convolution
    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):  # ch_in, ch_out, kernel, stride, padding, groups
        super().__init__()
        self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p), groups=g, bias=False)
        self.bn = nn.BatchNorm2d(c2)
        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)))

class Bottleneck(nn.Module):
    # Standard bottleneck
    def __init__(self, c1, c2, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, shortcut, groups, expansion
        super().__init__()
        c_ = int(c2 * e)  # hidden channels
        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))

class C3(nn.Module):
    # CSP Bottleneck with 3 convolutions
    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
        super().__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = Conv(c1, c_, 1, 1)
        self.cv3 = Conv(2 * c_, c2, 1)  # act=FReLU(c2)
        self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))

    def forward(self, x):
        return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(x)), dim=1))
    
class SPPF(nn.Module):
    # Spatial Pyramid Pooling - Fast (SPPF) layer for YOLOv5 by Glenn Jocher
    def __init__(self, c1, c2, k=5):  # equivalent to SPP(k=(5, 9, 13))
        super().__init__()
        c_ = c1 // 2  # hidden channels
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = Conv(c_ * 4, c2, 1, 1)
        self.m = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)

    def forward(self, x):
        x = self.cv1(x)
        with warnings.catch_warnings():
            warnings.simplefilter('ignore')  # suppress torch 1.9.0 max_pool2d() warning
            y1 = self.m(x)
            y2 = self.m(y1)
            return self.cv2(torch.cat([x, y1, y2, self.m(y2)], 1))
"""
这个是YOLOv5, 6.0版本的主干网络，这里进行复现
（注：有部分删改，详细讲解将在后续进行展开）
"""
class YOLOv5_backbone(nn.Module):
    def __init__(self):
        super(YOLOv5_backbone, self).__init__()
        
        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   = SPPF(1024, 1024, 5)
        
        # 全连接网络层，用于分类
        self.classifier = nn.Sequential(
            nn.Linear(in_features=65536, out_features=100),
            nn.ReLU(),
            nn.Linear(in_features=100, out_features=4)
        )
        
    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(x)
        
        x = torch.flatten(x, start_dim=1)
        x = self.classifier(x)

        return x

device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))
    
model = YOLOv5_backbone().to(device)
model

代码输出：（注：太长了，截取部分展示）

2.2 查看模型详情

# 统计模型参数量以及其他指标
import torchsummary as summary
summary.summary(model, (3, 224, 224))

代码输出：（注：太长了，截取部分展示）

3. 训练模型

3.1 编写训练函数

# 训练循环
def train(dataloader, model, loss_fn, optimizer):
    size = len(dataloader.dataset)  # 训练集的大小，一共60000张图片
    num_batches = len(dataloader)   # 批次数目，1875（60000/32）

    train_loss, train_acc = 0, 0  # 初始化训练损失和正确率
    
    for X, y in dataloader:  # 获取图片及其标签
        X, y = X.to(device), y.to(device)
        
        # 计算预测误差
        pred = model(X)          # 网络输出
        loss = loss_fn(pred, y)  # 计算网络输出和真实值之间的差距，targets为真实值，计算二者差值即为损失
        
        # 反向传播
        optimizer.zero_grad()  # grad属性归零
        loss.backward()        # 反向传播
        optimizer.step()       # 每一步自动更新
        
        # 记录acc与loss
        train_acc  += (pred.argmax(1) == y).type(torch.float).sum().item()
        train_loss += loss.item()
            
    train_acc  /= size
    train_loss /= num_batches

    return train_acc, train_loss

3.2 编写测试函数

def test (dataloader, model, loss_fn):
    size        = len(dataloader.dataset)  # 测试集的大小，一共10000张图片
    num_batches = len(dataloader)          # 批次数目，313（10000/32=312.5，向上取整）
    test_loss, test_acc = 0, 0
    
    # 当不进行训练时，停止梯度更新，节省计算内存消耗
    with torch.no_grad():
        for imgs, target in dataloader:
            imgs, target = imgs.to(device), target.to(device)
            
            # 计算loss
            target_pred = model(imgs)
            loss        = loss_fn(target_pred, target)
            
            test_loss += loss.item()
            test_acc  += (target_pred.argmax(1) == target).type(torch.float).sum().item()

    test_acc  /= size
    test_loss /= num_batches

    return test_acc, test_loss

3.3 正式训练

import copy

optimizer  = torch.optim.Adam(model.parameters(), lr= 1e-4)
loss_fn    = nn.CrossEntropyLoss() # 创建损失函数

epochs     = 60

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

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

for 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_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']
    
    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_9.pth'  # 保存的参数文件名
torch.save(model.state_dict(), PATH)

print('Done')

代码输出：

Epoch: 1,duration:14516ms, Train_acc:51.9%, Train_loss:1.178, Test_acc:59.1%, Test_loss:0.938, Lr:1.00E-04
Epoch: 2,duration:11299ms, Train_acc:65.0%, Train_loss:0.848, Test_acc:68.9%, Test_loss:0.592, Lr:1.00E-04
Epoch: 3,duration:11242ms, Train_acc:66.0%, Train_loss:0.772, Test_acc:79.6%, Test_loss:0.455, Lr:1.00E-04
Epoch: 4,duration:11300ms, Train_acc:71.0%, Train_loss:0.701, Test_acc:71.6%, Test_loss:0.561, Lr:1.00E-04
Epoch: 5,duration:11248ms, Train_acc:75.8%, Train_loss:0.629, Test_acc:75.1%, Test_loss:0.633, Lr:1.00E-04
Epoch: 6,duration:11275ms, Train_acc:80.3%, Train_loss:0.512, Test_acc:83.1%, Test_loss:0.455, Lr:1.00E-04
Epoch: 7,duration:11172ms, Train_acc:82.3%, Train_loss:0.476, Test_acc:84.4%, Test_loss:0.377, Lr:1.00E-04
Epoch: 8,duration:11269ms, Train_acc:85.3%, Train_loss:0.392, Test_acc:82.2%, Test_loss:0.373, Lr:1.00E-04
Epoch: 9,duration:11182ms, Train_acc:86.3%, Train_loss:0.355, Test_acc:80.4%, Test_loss:0.505, Lr:1.00E-04
Epoch:10,duration:11316ms, Train_acc:89.2%, Train_loss:0.280, Test_acc:88.4%, Test_loss:0.253, Lr:1.00E-04
Epoch:11,duration:11279ms, Train_acc:86.0%, Train_loss:0.355, Test_acc:85.8%, Test_loss:0.307, Lr:1.00E-04
Epoch:12,duration:11366ms, Train_acc:88.4%, Train_loss:0.273, Test_acc:85.8%, Test_loss:0.311, Lr:1.00E-04
Epoch:13,duration:11298ms, Train_acc:91.6%, Train_loss:0.252, Test_acc:90.7%, Test_loss:0.253, Lr:1.00E-04
Epoch:14,duration:11232ms, Train_acc:92.6%, Train_loss:0.223, Test_acc:91.1%, Test_loss:0.259, Lr:1.00E-04
Epoch:15,duration:11270ms, Train_acc:90.6%, Train_loss:0.237, Test_acc:91.1%, Test_loss:0.308, Lr:1.00E-04
Epoch:16,duration:11526ms, Train_acc:92.9%, Train_loss:0.178, Test_acc:96.4%, Test_loss:0.118, Lr:1.00E-04
Epoch:17,duration:11227ms, Train_acc:95.9%, Train_loss:0.095, Test_acc:93.3%, Test_loss:0.192, Lr:1.00E-04
Epoch:18,duration:11386ms, Train_acc:96.7%, Train_loss:0.100, Test_acc:90.7%, Test_loss:0.285, Lr:1.00E-04
Epoch:19,duration:11311ms, Train_acc:93.7%, Train_loss:0.165, Test_acc:88.4%, Test_loss:0.318, Lr:1.00E-04
Epoch:20,duration:12990ms, Train_acc:95.4%, Train_loss:0.122, Test_acc:93.8%, Test_loss:0.175, Lr:1.00E-04
Epoch:21,duration:13676ms, Train_acc:98.2%, Train_loss:0.063, Test_acc:93.8%, Test_loss:0.224, Lr:1.00E-04
Epoch:22,duration:13579ms, Train_acc:95.8%, Train_loss:0.128, Test_acc:93.3%, Test_loss:0.216, Lr:1.00E-04
Epoch:23,duration:13570ms, Train_acc:97.9%, Train_loss:0.048, Test_acc:93.8%, Test_loss:0.159, Lr:1.00E-04
Epoch:24,duration:13574ms, Train_acc:97.2%, Train_loss:0.060, Test_acc:89.3%, Test_loss:0.332, Lr:1.00E-04
Epoch:25,duration:13452ms, Train_acc:96.4%, Train_loss:0.109, Test_acc:94.2%, Test_loss:0.168, Lr:1.00E-04
Epoch:26,duration:13688ms, Train_acc:95.0%, Train_loss:0.145, Test_acc:88.9%, Test_loss:0.378, Lr:1.00E-04
Epoch:27,duration:12939ms, Train_acc:97.8%, Train_loss:0.064, Test_acc:95.1%, Test_loss:0.125, Lr:1.00E-04
Epoch:28,duration:13189ms, Train_acc:99.0%, Train_loss:0.027, Test_acc:95.6%, Test_loss:0.094, Lr:1.00E-04
Epoch:29,duration:13725ms, Train_acc:98.2%, Train_loss:0.049, Test_acc:94.7%, Test_loss:0.116, Lr:1.00E-04
Epoch:30,duration:13993ms, Train_acc:96.3%, Train_loss:0.125, Test_acc:90.2%, Test_loss:0.374, Lr:1.00E-04
Epoch:31,duration:13697ms, Train_acc:95.8%, Train_loss:0.120, Test_acc:94.7%, Test_loss:0.168, Lr:1.00E-04
Epoch:32,duration:13256ms, Train_acc:97.3%, Train_loss:0.081, Test_acc:89.8%, Test_loss:0.320, Lr:1.00E-04
Epoch:33,duration:13659ms, Train_acc:97.8%, Train_loss:0.058, Test_acc:94.2%, Test_loss:0.203, Lr:1.00E-04
Epoch:34,duration:12981ms, Train_acc:98.2%, Train_loss:0.048, Test_acc:88.9%, Test_loss:0.302, Lr:1.00E-04
Epoch:35,duration:13278ms, Train_acc:98.0%, Train_loss:0.050, Test_acc:93.3%, Test_loss:0.180, Lr:1.00E-04
Epoch:36,duration:13643ms, Train_acc:98.6%, Train_loss:0.041, Test_acc:93.8%, Test_loss:0.215, Lr:1.00E-04
Epoch:37,duration:12959ms, Train_acc:99.4%, Train_loss:0.016, Test_acc:95.1%, Test_loss:0.135, Lr:1.00E-04
Epoch:38,duration:13252ms, Train_acc:99.4%, Train_loss:0.016, Test_acc:93.3%, Test_loss:0.239, Lr:1.00E-04
Epoch:39,duration:13256ms, Train_acc:96.7%, Train_loss:0.104, Test_acc:88.4%, Test_loss:0.414, Lr:1.00E-04
Epoch:40,duration:13169ms, Train_acc:97.7%, Train_loss:0.071, Test_acc:92.4%, Test_loss:0.220, Lr:1.00E-04
Epoch:41,duration:12490ms, Train_acc:99.2%, Train_loss:0.026, Test_acc:94.7%, Test_loss:0.138, Lr:1.00E-04
Epoch:42,duration:11181ms, Train_acc:98.7%, Train_loss:0.042, Test_acc:92.0%, Test_loss:0.246, Lr:1.00E-04
Epoch:43,duration:11210ms, Train_acc:98.8%, Train_loss:0.045, Test_acc:90.7%, Test_loss:0.299, Lr:1.00E-04
Epoch:44,duration:11399ms, Train_acc:99.1%, Train_loss:0.028, Test_acc:90.7%, Test_loss:0.316, Lr:1.00E-04
Epoch:45,duration:11255ms, Train_acc:97.6%, Train_loss:0.069, Test_acc:91.1%, Test_loss:0.251, Lr:1.00E-04
Epoch:46,duration:11540ms, Train_acc:98.4%, Train_loss:0.050, Test_acc:92.0%, Test_loss:0.282, Lr:1.00E-04
Epoch:47,duration:14299ms, Train_acc:97.7%, Train_loss:0.074, Test_acc:90.2%, Test_loss:0.289, Lr:1.00E-04
Epoch:48,duration:13733ms, Train_acc:98.2%, Train_loss:0.045, Test_acc:94.2%, Test_loss:0.196, Lr:1.00E-04
Epoch:49,duration:13832ms, Train_acc:98.8%, Train_loss:0.029, Test_acc:92.4%, Test_loss:0.248, Lr:1.00E-04
Epoch:50,duration:13235ms, Train_acc:98.9%, Train_loss:0.029, Test_acc:93.8%, Test_loss:0.182, Lr:1.00E-04
Epoch:51,duration:13804ms, Train_acc:99.0%, Train_loss:0.030, Test_acc:89.3%, Test_loss:0.328, Lr:1.00E-04
Epoch:52,duration:13920ms, Train_acc:98.2%, Train_loss:0.045, Test_acc:91.6%, Test_loss:0.235, Lr:1.00E-04
Epoch:53,duration:14425ms, Train_acc:99.1%, Train_loss:0.021, Test_acc:84.9%, Test_loss:0.693, Lr:1.00E-04
Epoch:54,duration:14777ms, Train_acc:97.0%, Train_loss:0.109, Test_acc:88.4%, Test_loss:0.540, Lr:1.00E-04
Epoch:55,duration:14075ms, Train_acc:97.4%, Train_loss:0.067, Test_acc:89.8%, Test_loss:0.365, Lr:1.00E-04
Epoch:56,duration:13698ms, Train_acc:99.4%, Train_loss:0.016, Test_acc:91.6%, Test_loss:0.337, Lr:1.00E-04
Epoch:57,duration:13465ms, Train_acc:98.7%, Train_loss:0.040, Test_acc:96.0%, Test_loss:0.189, Lr:1.00E-04
Epoch:58,duration:14171ms, Train_acc:99.0%, Train_loss:0.028, Test_acc:91.6%, Test_loss:0.208, Lr:1.00E-04
Epoch:59,duration:13446ms, Train_acc:99.9%, Train_loss:0.009, Test_acc:93.8%, Test_loss:0.213, Lr:1.00E-04
Epoch:60,duration:13758ms, Train_acc:98.7%, Train_loss:0.042, Test_acc:95.1%, Test_loss:0.160, Lr:1.00E-04
Done

4. 结果可视化

4.1 Loss 与 Accuracy 图

import matplotlib.pyplot as plt
#隐藏警告
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

epochs_range = range(epochs)

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 and Validation 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 and Validation Loss')
plt.show()

代码输出：

4.2 模型评估

best_model.eval()
epoch_test_acc, epoch_test_loss = test(test_dl, best_model, loss_fn)
epoch_test_acc, epoch_test_loss

代码输出：

0.9644444444444444 0.11823589946288848

个人碎碎念：

这周通过手动搭建包含Backbone模块的模型，更难了，需要更多的时间来消化了……

帅先森！

关注

4
点赞
踩
7

收藏

觉得还不错? 一键收藏
0
评论
复制链接

分享到 QQ

分享到新浪微博

扫一扫

专栏目录