P5 运动鞋识别

最新推荐文章于 2024-05-23 20:56:45 发布

月亮泡面=

最新推荐文章于 2024-05-23 20:56:45 发布

阅读量857

点赞数 21

文章标签：学习

本文链接：https://blog.csdn.net/weixin_45103399/article/details/135777217

版权

🍨 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 | 接辅导、项目定制
🚀 文章来源：K同学的学习圈子

我的环境：

语言环境：python 3.8
编译器：jupyter notebook
深度学习环境：Pytorch

torch == 2.1.0+cpu

torchvision == 0.16.0+cpu
要求：

设置动态学习率。
调整代码使测试集accuracy到达84%。

一、准备工作

1. 导入库函数

import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
from torchvision import transforms, datasets

import os,PIL,pathlib,random

2.加载数据

● 第一步：使用pathlib.Path()函数将字符串类型的文件夹路径转换为pathlib.Path对象。

● 第二步：使用glob()方法获取data_dir路径下的所有文件路径，并以列表形式存储在data_paths中。

● 第三步：通过split()函数对data_paths中的每个文件路径执行分割操作，获得各个文件所属的类别名称，并存储在classNames中

● 第四步：打印classNames列表，显示每个文件所属的类别名称。

data_dir = "D:/BaiduNetdiskDownload/datasets/t5"
data_dir = pathlib.Path(data_dir)
data_paths = list(data_dir.glob('*/'))
classNames = [str(path).split("\\")[4] for path in data_paths]
classNames

['test', 'train']

3. 数据加载（设置batchsize和取样等功能）

# 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.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。
])

test_transform = transforms.Compose([
    transforms.Resize([224, 224]),  # 将输入图片resize成统一尺寸
    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。
])

train_dataset = datasets.ImageFolder("D:/BaiduNetdiskDownload/datasets/t5/train",transform=train_transforms)
test_dataset  = datasets.ImageFolder("D:/BaiduNetdiskDownload/datasets/t5/test/",transform=train_transforms)

train_dataset.class_to_idx

{'adidas': 0, 'nike': 1}

batch_size = 32

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

Shape of X [N, C, H, W]:  torch.Size([32, 3, 224, 224])
Shape of y:  torch.Size([32]) torch.int64

二、CNN网络配置、编译、训练

在这里插入图片描述

1. 模型建立

import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.conv1=nn.Sequential(
            nn.Conv2d(3, 12, kernel_size=5, padding=0), # 12*220*220
            nn.BatchNorm2d(12),
            nn.ReLU())
        
        self.conv2=nn.Sequential(
            nn.Conv2d(12, 12, kernel_size=5, padding=0), # 12*216*216
            nn.BatchNorm2d(12),
            nn.ReLU())
        
        self.pool3=nn.Sequential(
            nn.MaxPool2d(2))                              # 12*108*108
        
        self.conv4=nn.Sequential(
            nn.Conv2d(12, 24, kernel_size=5, padding=0), # 24*104*104
            nn.BatchNorm2d(24),
            nn.ReLU())
        
        self.conv5=nn.Sequential(
            nn.Conv2d(24, 24, kernel_size=5, padding=0), # 24*100*100
            nn.BatchNorm2d(24),
            nn.ReLU())
        
        self.pool6=nn.Sequential(
            nn.MaxPool2d(2))                              # 24*50*50

        self.dropout = nn.Sequential(
            nn.Dropout(0.2))
        
        self.fc=nn.Sequential(
            nn.Linear(24*50*50, len(classNames)))
        
    def forward(self, x):
        
        batch_size = x.size(0)
        x = self.conv1(x)  # 卷积-BN-激活
        x = self.conv2(x)  # 卷积-BN-激活
        x = self.pool3(x)  # 池化
        x = self.conv4(x)  # 卷积-BN-激活
        x = self.conv5(x)  # 卷积-BN-激活
        x = self.pool6(x)  # 池化
        x = self.dropout(x)
        x = x.view(batch_size, -1)  # flatten 变成全连接网络需要的输入 (batch, 24*50*50) ==> (batch, -1), -1 此处自动算出的是24*50*50
        x = self.fc(x)
       
        return x

model = Model()
model

Model(
  (conv1): Sequential(
    (0): Conv2d(3, 12, kernel_size=(5, 5), stride=(1, 1))
    (1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (conv2): Sequential(
    (0): Conv2d(12, 12, kernel_size=(5, 5), stride=(1, 1))
    (1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (pool3): Sequential(
    (0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (conv4): Sequential(
    (0): Conv2d(12, 24, kernel_size=(5, 5), stride=(1, 1))
    (1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (conv5): Sequential(
    (0): Conv2d(24, 24, kernel_size=(5, 5), stride=(1, 1))
    (1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (pool6): Sequential(
    (0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (dropout): Sequential(
    (0): Dropout(p=0.2, inplace=False)
  )
  (fc): Sequential(
    (0): Linear(in_features=60000, out_features=2, bias=True)
  )
)

import torchinfo
torchinfo.summary(model) #显示模型信息

=================================================================
Layer (type:depth-idx)                   Param #
=================================================================
Model                                    --
├─Sequential: 1-1                        --
│    └─Conv2d: 2-1                       912
│    └─BatchNorm2d: 2-2                  24
│    └─ReLU: 2-3                         --
├─Sequential: 1-2                        --
│    └─Conv2d: 2-4                       3,612
│    └─BatchNorm2d: 2-5                  24
│    └─ReLU: 2-6                         --
├─Sequential: 1-3                        --
│    └─MaxPool2d: 2-7                    --
├─Sequential: 1-4                        --
│    └─Conv2d: 2-8                       7,224
│    └─BatchNorm2d: 2-9                  48
│    └─ReLU: 2-10                        --
├─Sequential: 1-5                        --
│    └─Conv2d: 2-11                      14,424
│    └─BatchNorm2d: 2-12                 48
│    └─ReLU: 2-13                        --
├─Sequential: 1-6                        --
│    └─MaxPool2d: 2-14                   --
├─Sequential: 1-7                        --
│    └─Dropout: 2-15                     --
├─Sequential: 1-8                        --
│    └─Linear: 2-16                      120,002
=================================================================
Total params: 146,318
Trainable params: 146,318
Non-trainable params: 0
=================================================================

2. 训练函数

2.1 训练函数

optimizer.zero_grad(),梯度清零。
loss.backward(),反向传播计算每个w的梯度值。
optimizer.step(),梯度下降法更新参数值。

# 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

train_acc和loss的计算原理：

pred.argmax(1)返回预测结果pred行最大值的索引，每行是表示一个样本的预测概率分布。==y表示判断预测是否正确。
.type(torch.float)将判断结果转为浮点类型可以进行求和。
.sum()对预测结果的正误进行求和。
.item()将求和结果转为标量值便于输出。

2.2 测试函数

去掉了训练函数中梯度下降和权重更新的步骤。

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

2.3 设置动态学习率

def adjust_learning_rate(optimizer, epoch, start_lr):
    # 每 2 个epoch衰减到原来的 0.98
    lr = start_lr * (0.92 ** (epoch // 2))
    for param_group in optimizer.param_groups:
        param_group['lr'] = lr

learn_rate = 1e-4 # 初始学习率
optimizer  = torch.optim.SGD(model.parameters(), lr=learn_rate)
#opt = torch.optim.Adam(model.parameters(), lr=learn_rate)

# # 调用官方动态学习率接口时使用
# lambda1 = lambda epoch: (0.92 ** (epoch // 2)
# optimizer = torch.optim.SGD(model.parameters(), lr=learn_rate)
# scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=lambda1) #选定调整方法

2.4 模型训练

model.train(),用于训练，启用BN层和dropout。
model.eval(),用于测试，不启用BN层和dropout。

第一次训练时忘记了加入学习率更新，得到的结果：

在这里插入图片描述

加入学习率更新后：

loss_fn    = nn.CrossEntropyLoss() # 创建损失函数
epochs     = 50
train_loss = []
train_acc  = []
test_loss  = []
test_acc   = []
best_test_acc = 0
PATH = './model.pth'
for epoch in range(epochs):
    # 更新学习率（使用自定义学习率时使用）
    adjust_learning_rate(optimizer, epoch, learn_rate)
    
    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)
    
    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))
    if best_test_acc < epoch_test_acc:
        best_test_acc = epoch_test_acc
        torch.save(model.state_dict(),PATH)
        print('save model')
print('Done')

Epoch: 1, Train_acc:54.6%, Train_loss:0.716, Test_acc:51.3%, Test_loss:0.695, Lr:1.00E-04
save model
Epoch: 2, Train_acc:63.5%, Train_loss:0.660, Test_acc:61.8%, Test_loss:0.635, Lr:1.00E-04
save model
Epoch: 3, Train_acc:67.5%, Train_loss:0.602, Test_acc:72.4%, Test_loss:0.571, Lr:9.20E-05
save model
Epoch: 4, Train_acc:73.3%, Train_loss:0.550, Test_acc:72.4%, Test_loss:0.547, Lr:9.20E-05
Epoch: 5, Train_acc:74.1%, Train_loss:0.546, Test_acc:76.3%, Test_loss:0.544, Lr:8.46E-05
save model
Epoch: 6, Train_acc:77.7%, Train_loss:0.504, Test_acc:73.7%, Test_loss:0.550, Lr:8.46E-05
Epoch: 7, Train_acc:79.7%, Train_loss:0.485, Test_acc:76.3%, Test_loss:0.477, Lr:7.79E-05
Epoch: 8, Train_acc:81.9%, Train_loss:0.462, Test_acc:80.3%, Test_loss:0.502, Lr:7.79E-05
save model
Epoch: 9, Train_acc:84.1%, Train_loss:0.441, Test_acc:78.9%, Test_loss:0.440, Lr:7.16E-05
Epoch:10, Train_acc:83.7%, Train_loss:0.426, Test_acc:82.9%, Test_loss:0.460, Lr:7.16E-05
save model
Epoch:11, Train_acc:84.1%, Train_loss:0.412, Test_acc:81.6%, Test_loss:0.458, Lr:6.59E-05
Epoch:12, Train_acc:87.1%, Train_loss:0.392, Test_acc:81.6%, Test_loss:0.458, Lr:6.59E-05
Epoch:13, Train_acc:85.5%, Train_loss:0.391, Test_acc:81.6%, Test_loss:0.446, Lr:6.06E-05
Epoch:14, Train_acc:88.6%, Train_loss:0.368, Test_acc:81.6%, Test_loss:0.462, Lr:6.06E-05
Epoch:15, Train_acc:88.8%, Train_loss:0.363, Test_acc:81.6%, Test_loss:0.400, Lr:5.58E-05
Epoch:16, Train_acc:88.0%, Train_loss:0.356, Test_acc:81.6%, Test_loss:0.402, Lr:5.58E-05
Epoch:17, Train_acc:90.6%, Train_loss:0.338, Test_acc:82.9%, Test_loss:0.404, Lr:5.13E-05
Epoch:18, Train_acc:91.4%, Train_loss:0.322, Test_acc:81.6%, Test_loss:0.447, Lr:5.13E-05
Epoch:19, Train_acc:91.2%, Train_loss:0.327, Test_acc:81.6%, Test_loss:0.398, Lr:4.72E-05
Epoch:20, Train_acc:91.2%, Train_loss:0.320, Test_acc:82.9%, Test_loss:0.431, Lr:4.72E-05
Epoch:21, Train_acc:90.8%, Train_loss:0.335, Test_acc:81.6%, Test_loss:0.418, Lr:4.34E-05
Epoch:22, Train_acc:91.6%, Train_loss:0.316, Test_acc:84.2%, Test_loss:0.378, Lr:4.34E-05
save model
Epoch:23, Train_acc:92.8%, Train_loss:0.302, Test_acc:81.6%, Test_loss:0.372, Lr:4.00E-05
Epoch:24, Train_acc:92.8%, Train_loss:0.298, Test_acc:81.6%, Test_loss:0.422, Lr:4.00E-05
Epoch:25, Train_acc:93.2%, Train_loss:0.294, Test_acc:82.9%, Test_loss:0.375, Lr:3.68E-05
Epoch:26, Train_acc:93.4%, Train_loss:0.296, Test_acc:81.6%, Test_loss:0.347, Lr:3.68E-05
Epoch:27, Train_acc:95.0%, Train_loss:0.285, Test_acc:84.2%, Test_loss:0.363, Lr:3.38E-05
Epoch:28, Train_acc:93.4%, Train_loss:0.283, Test_acc:81.6%, Test_loss:0.386, Lr:3.38E-05
Epoch:29, Train_acc:93.0%, Train_loss:0.279, Test_acc:84.2%, Test_loss:0.398, Lr:3.11E-05
Epoch:30, Train_acc:94.2%, Train_loss:0.280, Test_acc:82.9%, Test_loss:0.398, Lr:3.11E-05
Epoch:31, Train_acc:93.4%, Train_loss:0.284, Test_acc:81.6%, Test_loss:0.381, Lr:2.86E-05
Epoch:32, Train_acc:95.4%, Train_loss:0.270, Test_acc:84.2%, Test_loss:0.377, Lr:2.86E-05
Epoch:33, Train_acc:93.8%, Train_loss:0.268, Test_acc:82.9%, Test_loss:0.396, Lr:2.63E-05
Epoch:34, Train_acc:96.0%, Train_loss:0.258, Test_acc:82.9%, Test_loss:0.403, Lr:2.63E-05
Epoch:35, Train_acc:94.8%, Train_loss:0.265, Test_acc:82.9%, Test_loss:0.384, Lr:2.42E-05
Epoch:36, Train_acc:95.8%, Train_loss:0.266, Test_acc:82.9%, Test_loss:0.405, Lr:2.42E-05
Epoch:37, Train_acc:94.8%, Train_loss:0.263, Test_acc:84.2%, Test_loss:0.362, Lr:2.23E-05
Epoch:38, Train_acc:95.4%, Train_loss:0.256, Test_acc:81.6%, Test_loss:0.425, Lr:2.23E-05
Epoch:39, Train_acc:96.4%, Train_loss:0.249, Test_acc:82.9%, Test_loss:0.398, Lr:2.05E-05
Epoch:40, Train_acc:96.2%, Train_loss:0.247, Test_acc:82.9%, Test_loss:0.390, Lr:2.05E-05
Epoch:41, Train_acc:96.4%, Train_loss:0.246, Test_acc:82.9%, Test_loss:0.368, Lr:1.89E-05
Epoch:42, Train_acc:96.6%, Train_loss:0.246, Test_acc:84.2%, Test_loss:0.388, Lr:1.89E-05
Epoch:43, Train_acc:95.6%, Train_loss:0.250, Test_acc:82.9%, Test_loss:0.406, Lr:1.74E-05
Epoch:44, Train_acc:96.4%, Train_loss:0.244, Test_acc:84.2%, Test_loss:0.361, Lr:1.74E-05
Epoch:45, Train_acc:95.4%, Train_loss:0.253, Test_acc:82.9%, Test_loss:0.378, Lr:1.60E-05
Epoch:46, Train_acc:96.2%, Train_loss:0.238, Test_acc:84.2%, Test_loss:0.354, Lr:1.60E-05
Epoch:47, Train_acc:95.6%, Train_loss:0.247, Test_acc:84.2%, Test_loss:0.425, Lr:1.47E-05
Epoch:48, Train_acc:94.4%, Train_loss:0.250, Test_acc:82.9%, Test_loss:0.383, Lr:1.47E-05
Epoch:49, Train_acc:97.0%, Train_loss:0.234, Test_acc:82.9%, Test_loss:0.356, Lr:1.35E-05
Epoch:50, Train_acc:94.8%, Train_loss:0.245, Test_acc:82.9%, Test_loss:0.361, Lr:1.35E-05
Done

三、训练结果可视化

1. loss and accuracy

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()

在这里插入图片描述

测试准确率最好的时候达到了84.2%，没有进一步提升。

2. 指定图片预测

torch.squeeze(input,dim = None)，进行维度压缩，去掉维度为1的维度，或去掉dim维度。
torch.unsqueeze(input,dim = int)，进行维度扩充，在dim = int的维度增加维数为1的维度。

from PIL import Image 

classes = list(test_dataset.class_to_idx)

def predict_one_image(image_path, model, transform, classes):
    
    test_img = Image.open(image_path).convert('RGB')
    # plt.imshow(test_img)  # 展示预测的图片

    test_img = transform(test_img)
    img = test_img.unsqueeze(0)
    
    model.eval()
    output = model(img)

    _,pred = torch.max(output,1)
    pred_class = classes[pred]
    print(f'预测结果是：{pred_class}')

# 加载保存的准确率最好的模型
loaded_model = model
loaded_model.load_state_dict(torch.load(PATH))
print(loaded_model)

Model(
  (conv1): Sequential(
    (0): Conv2d(3, 12, kernel_size=(5, 5), stride=(1, 1))
    (1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (conv2): Sequential(
    (0): Conv2d(12, 12, kernel_size=(5, 5), stride=(1, 1))
    (1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (pool3): Sequential(
    (0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (conv4): Sequential(
    (0): Conv2d(12, 24, kernel_size=(5, 5), stride=(1, 1))
    (1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (conv5): Sequential(
    (0): Conv2d(24, 24, kernel_size=(5, 5), stride=(1, 1))
    (1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (pool6): Sequential(
    (0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (dropout): Sequential(
    (0): Dropout(p=0.2, inplace=False)
  )
  (fc): Sequential(
    (0): Linear(in_features=60000, out_features=2, bias=True)
  )
)

#进行预测
predict_one_image(image_path='D:/BaiduNetdiskDownload/datasets/t5/test/adidas/5.jpg', 
                  model=loaded_model, 
                  transform=train_transforms, 
                  classes=classes)

预测结果是：adidas

准确率优化

修改优化器为Adam：

def adjust_learning_rate(optimizer, epoch, start_lr):
    # 每 2 个epoch衰减到原来的 0.98
    lr = start_lr * (0.92 ** (epoch // 2))
    for param_group in optimizer.param_groups:
        param_group['lr'] = lr

learn_rate = 1e-4 # 初始学习率
#optimizer  = torch.optim.SGD(model.parameters(), lr=learn_rate)
optimizer = torch.optim.Adam(model.parameters(), lr=learn_rate)

loss_fn    = nn.CrossEntropyLoss() # 创建损失函数
epochs     = 40
train_loss = []
train_acc  = []
test_loss  = []
test_acc   = []
best_test_acc = 0
PATH = './model.pth'
for epoch in range(epochs):
    # 更新学习率（使用自定义学习率时使用）
    adjust_learning_rate(optimizer, epoch, learn_rate)
    
    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)
    
    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))
    if best_test_acc < epoch_test_acc:
        best_test_acc = epoch_test_acc
        torch.save(model.state_dict(),PATH)
        print('save model')
print('Done')

Epoch: 1, Train_acc:66.1%, Train_loss:0.827, Test_acc:80.3%, Test_loss:0.399, Lr:1.00E-04
save model
Epoch: 2, Train_acc:80.5%, Train_loss:0.397, Test_acc:86.8%, Test_loss:0.368, Lr:1.00E-04
save model
Epoch: 3, Train_acc:92.2%, Train_loss:0.223, Test_acc:82.9%, Test_loss:0.360, Lr:9.20E-05
Epoch: 4, Train_acc:95.6%, Train_loss:0.172, Test_acc:82.9%, Test_loss:0.338, Lr:9.20E-05
Epoch: 5, Train_acc:98.2%, Train_loss:0.131, Test_acc:85.5%, Test_loss:0.299, Lr:8.46E-05
Epoch: 6, Train_acc:99.0%, Train_loss:0.114, Test_acc:85.5%, Test_loss:0.380, Lr:8.46E-05
Epoch: 7, Train_acc:99.2%, Train_loss:0.101, Test_acc:86.8%, Test_loss:0.311, Lr:7.79E-05
Epoch: 8, Train_acc:99.4%, Train_loss:0.085, Test_acc:89.5%, Test_loss:0.319, Lr:7.79E-05
save model
Epoch: 9, Train_acc:99.4%, Train_loss:0.069, Test_acc:85.5%, Test_loss:0.262, Lr:7.16E-05
Epoch:10, Train_acc:100.0%, Train_loss:0.069, Test_acc:89.5%, Test_loss:0.281, Lr:7.16E-05
Epoch:11, Train_acc:100.0%, Train_loss:0.054, Test_acc:89.5%, Test_loss:0.311, Lr:6.59E-05
Epoch:12, Train_acc:100.0%, Train_loss:0.047, Test_acc:89.5%, Test_loss:0.244, Lr:6.59E-05
Epoch:13, Train_acc:100.0%, Train_loss:0.046, Test_acc:85.5%, Test_loss:0.249, Lr:6.06E-05
Epoch:14, Train_acc:100.0%, Train_loss:0.043, Test_acc:88.2%, Test_loss:0.256, Lr:6.06E-05
Epoch:15, Train_acc:100.0%, Train_loss:0.041, Test_acc:82.9%, Test_loss:0.305, Lr:5.58E-05
Epoch:16, Train_acc:100.0%, Train_loss:0.036, Test_acc:89.5%, Test_loss:0.275, Lr:5.58E-05
Epoch:17, Train_acc:100.0%, Train_loss:0.031, Test_acc:86.8%, Test_loss:0.291, Lr:5.13E-05
Epoch:18, Train_acc:100.0%, Train_loss:0.031, Test_acc:86.8%, Test_loss:0.321, Lr:5.13E-05
Epoch:19, Train_acc:100.0%, Train_loss:0.029, Test_acc:88.2%, Test_loss:0.241, Lr:4.72E-05
Epoch:20, Train_acc:100.0%, Train_loss:0.027, Test_acc:89.5%, Test_loss:0.294, Lr:4.72E-05
Epoch:21, Train_acc:100.0%, Train_loss:0.030, Test_acc:86.8%, Test_loss:0.329, Lr:4.34E-05
Epoch:22, Train_acc:100.0%, Train_loss:0.027, Test_acc:88.2%, Test_loss:0.269, Lr:4.34E-05
Epoch:23, Train_acc:100.0%, Train_loss:0.026, Test_acc:88.2%, Test_loss:0.295, Lr:4.00E-05
Epoch:24, Train_acc:100.0%, Train_loss:0.023, Test_acc:85.5%, Test_loss:0.251, Lr:4.00E-05
Epoch:25, Train_acc:100.0%, Train_loss:0.020, Test_acc:89.5%, Test_loss:0.303, Lr:3.68E-05
Epoch:26, Train_acc:100.0%, Train_loss:0.020, Test_acc:86.8%, Test_loss:0.341, Lr:3.68E-05
Epoch:27, Train_acc:100.0%, Train_loss:0.020, Test_acc:88.2%, Test_loss:0.261, Lr:3.38E-05
Epoch:28, Train_acc:100.0%, Train_loss:0.021, Test_acc:89.5%, Test_loss:0.245, Lr:3.38E-05
Epoch:29, Train_acc:100.0%, Train_loss:0.022, Test_acc:86.8%, Test_loss:0.308, Lr:3.11E-05
Epoch:30, Train_acc:100.0%, Train_loss:0.023, Test_acc:88.2%, Test_loss:0.352, Lr:3.11E-05
Epoch:31, Train_acc:100.0%, Train_loss:0.019, Test_acc:89.5%, Test_loss:0.271, Lr:2.86E-05
Epoch:32, Train_acc:100.0%, Train_loss:0.017, Test_acc:88.2%, Test_loss:0.287, Lr:2.86E-05
Epoch:33, Train_acc:100.0%, Train_loss:0.016, Test_acc:89.5%, Test_loss:0.263, Lr:2.63E-05
Epoch:34, Train_acc:100.0%, Train_loss:0.018, Test_acc:89.5%, Test_loss:0.271, Lr:2.63E-05
Epoch:35, Train_acc:100.0%, Train_loss:0.016, Test_acc:88.2%, Test_loss:0.309, Lr:2.42E-05
Epoch:36, Train_acc:100.0%, Train_loss:0.016, Test_acc:89.5%, Test_loss:0.273, Lr:2.42E-05
Epoch:37, Train_acc:100.0%, Train_loss:0.017, Test_acc:88.2%, Test_loss:0.257, Lr:2.23E-05
Epoch:38, Train_acc:100.0%, Train_loss:0.015, Test_acc:88.2%, Test_loss:0.362, Lr:2.23E-05
Epoch:39, Train_acc:100.0%, Train_loss:0.014, Test_acc:88.2%, Test_loss:0.282, Lr:2.05E-05
Epoch:40, Train_acc:100.0%, Train_loss:0.014, Test_acc:89.5%, Test_loss:0.249, Lr:2.05E-05
Done

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()

在这里插入图片描述

更换了Adam优化器后，在10轮训练就取得了稳定的结果，最好的表现是89.5%。优于SGD优化器。

#进行预测
predict_one_image(image_path='D:/BaiduNetdiskDownload/datasets/t5/test/adidas/5.jpg', 
                  model=model, 
                  transform=train_transforms, 
                  classes=classes)