深度学习实验--第P5周：Pytorch实现运动鞋识别

原创已于 2025-07-09 17:26:11 修改 · 287 阅读

3 ·

CC 4.0 BY-SA版权

文章标签：

#深度学习 #pytorch #人工智能

于 2025-06-23 17:31:54 首次发布

AI学习专栏收录该内容

6 篇文章

订阅专栏

部署运行你感兴趣的模型镜像

🍨 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
我的环境：

语言环境：Python3.12
编译器：PyCharm
深度学习环境：
torch==1.12.1+cu113
torchvision==0.13.1+cu113

一、实验

1、目的：

学习如何设置动态学习率，如何保存模型

2、总结：

1）训练完成后保存模型；

2）学习通过不同方式来设置学习率，观察训练模型准确率差别；

方法1：自定义函数调用LambdaLR：94.8%

方法2：调用LambdaLR：97.2%,

方法3：等间隔动态调整调用StepLR：88.4%

方法4：自定义的函数调用LambdaLR（同1、2 lambda1参数）：86.3%

调用LambdaLR准确率最高，保存最高准确率函数。

自定义设置学习率

def adjust_learning_rate(optimizer, epoch, start_lr):
    # 每 2 个epoch衰减到原来的 0.92
    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)

调用官方动态学习率接口-LambdaLR

 # 调用官方动态学习率接口时使用LambdaLR
 learn_rate = 1e-4 # 初始学习率
 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

调用官方动态学习率接口-StepLR

# # 调用官方动态学习率接口时使用StepLR
optimizer = torch.optim.SGD(model.parameters(), lr=0.001 )
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=5, gamma=0.1)

调用官方动态学习率接口-MultiStepLR

# # 调用官方动态学习率接口时使用MultiStepLR
optimizer = torch.optim.SGD(model.parameters(), lr=0.001 )
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer,
                                                 milestones=[2,6,15], #调整学习率的epoch数
                                                 gamma=0.1)

3、结果：

自定义设置学习率

D:\Programs\Python\Python39\python.exe D:\PycharmProjects\pythonProject\P5\main.py 
cpu
['test', 'train']
{'adidas': 0, 'nike': 1}
Shape of X [N, C, H, W]:  torch.Size([32, 3, 224, 224])
Shape of y:  torch.Size([32]) torch.int64
Using cpu device
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)
  )
)
Epoch: 1, Train_acc:53.2%, Train_loss:0.741, Test_acc:51.3%, Test_loss:0.688, Lr:1.00E-04
Epoch: 2, Train_acc:62.0%, Train_loss:0.662, Test_acc:63.2%, Test_loss:0.651, Lr:1.00E-04
Epoch: 3, Train_acc:68.5%, Train_loss:0.607, Test_acc:67.1%, Test_loss:0.612, Lr:9.20E-05
Epoch: 4, Train_acc:71.9%, Train_loss:0.562, Test_acc:69.7%, Test_loss:0.620, Lr:9.20E-05
Epoch: 5, Train_acc:76.5%, Train_loss:0.518, Test_acc:69.7%, Test_loss:0.580, Lr:8.46E-05
Epoch: 6, Train_acc:75.7%, Train_loss:0.501, Test_acc:71.1%, Test_loss:0.597, Lr:8.46E-05
Epoch: 7, Train_acc:77.9%, Train_loss:0.484, Test_acc:71.1%, Test_loss:0.572, Lr:7.79E-05
Epoch: 8, Train_acc:80.5%, Train_loss:0.467, Test_acc:71.1%, Test_loss:0.578, Lr:7.79E-05
Epoch: 9, Train_acc:81.1%, Train_loss:0.440, Test_acc:72.4%, Test_loss:0.561, Lr:7.16E-05
Epoch:10, Train_acc:84.5%, Train_loss:0.406, Test_acc:72.4%, Test_loss:0.570, Lr:7.16E-05
Epoch:11, Train_acc:85.3%, Train_loss:0.411, Test_acc:72.4%, Test_loss:0.513, Lr:6.59E-05
Epoch:12, Train_acc:85.3%, Train_loss:0.400, Test_acc:72.4%, Test_loss:0.601, Lr:6.59E-05
Epoch:13, Train_acc:87.6%, Train_loss:0.374, Test_acc:75.0%, Test_loss:0.514, Lr:6.06E-05
Epoch:14, Train_acc:85.3%, Train_loss:0.390, Test_acc:72.4%, Test_loss:0.531, Lr:6.06E-05
Epoch:15, Train_acc:88.4%, Train_loss:0.369, Test_acc:73.7%, Test_loss:0.533, Lr:5.58E-05
Epoch:16, Train_acc:87.6%, Train_loss:0.359, Test_acc:75.0%, Test_loss:0.500, Lr:5.58E-05
Epoch:17, Train_acc:88.8%, Train_loss:0.349, Test_acc:72.4%, Test_loss:0.475, Lr:5.13E-05
Epoch:18, Train_acc:88.6%, Train_loss:0.357, Test_acc:73.7%, Test_loss:0.482, Lr:5.13E-05
Epoch:19, Train_acc:90.8%, Train_loss:0.338, Test_acc:73.7%, Test_loss:0.461, Lr:4.72E-05
Epoch:20, Train_acc:90.0%, Train_loss:0.336, Test_acc:73.7%, Test_loss:0.545, Lr:4.72E-05
Epoch:21, Train_acc:90.6%, Train_loss:0.318, Test_acc:75.0%, Test_loss:0.538, Lr:4.34E-05
Epoch:22, Train_acc:93.2%, Train_loss:0.315, Test_acc:76.3%, Test_loss:0.493, Lr:4.34E-05
Epoch:23, Train_acc:93.2%, Train_loss:0.306, Test_acc:73.7%, Test_loss:0.490, Lr:4.00E-05
Epoch:24, Train_acc:92.2%, Train_loss:0.305, Test_acc:73.7%, Test_loss:0.471, Lr:4.00E-05
Epoch:25, Train_acc:92.2%, Train_loss:0.307, Test_acc:76.3%, Test_loss:0.547, Lr:3.68E-05
Epoch:26, Train_acc:92.2%, Train_loss:0.307, Test_acc:75.0%, Test_loss:0.506, Lr:3.68E-05
Epoch:27, Train_acc:94.0%, Train_loss:0.294, Test_acc:76.3%, Test_loss:0.493, Lr:3.38E-05
Epoch:28, Train_acc:94.2%, Train_loss:0.295, Test_acc:76.3%, Test_loss:0.487, Lr:3.38E-05
Epoch:29, Train_acc:93.6%, Train_loss:0.286, Test_acc:76.3%, Test_loss:0.470, Lr:3.11E-05
Epoch:30, Train_acc:94.2%, Train_loss:0.284, Test_acc:76.3%, Test_loss:0.508, Lr:3.11E-05
Epoch:31, Train_acc:92.8%, Train_loss:0.286, Test_acc:76.3%, Test_loss:0.518, Lr:2.86E-05
Epoch:32, Train_acc:95.4%, Train_loss:0.287, Test_acc:76.3%, Test_loss:0.541, Lr:2.86E-05
Epoch:33, Train_acc:93.6%, Train_loss:0.281, Test_acc:76.3%, Test_loss:0.483, Lr:2.63E-05
Epoch:34, Train_acc:94.0%, Train_loss:0.273, Test_acc:76.3%, Test_loss:0.466, Lr:2.63E-05
Epoch:35, Train_acc:94.8%, Train_loss:0.273, Test_acc:76.3%, Test_loss:0.468, Lr:2.42E-05
Epoch:36, Train_acc:93.4%, Train_loss:0.275, Test_acc:76.3%, Test_loss:0.448, Lr:2.42E-05
Epoch:37, Train_acc:93.4%, Train_loss:0.270, Test_acc:76.3%, Test_loss:0.453, Lr:2.23E-05
Epoch:38, Train_acc:94.6%, Train_loss:0.272, Test_acc:76.3%, Test_loss:0.466, Lr:2.23E-05
Epoch:39, Train_acc:94.2%, Train_loss:0.281, Test_acc:76.3%, Test_loss:0.486, Lr:2.05E-05
Epoch:40, Train_acc:94.8%, Train_loss:0.269, Test_acc:76.3%, Test_loss:0.471, Lr:2.05E-05
Done
预测结果是：adidas

进程已结束，退出代码为 0

调用官方动态学习率接口-LambdaLR

D:\Programs\Python\Python39\python.exe D:\PycharmProjects\pythonProject\P5\main.py 
cpu
['test', 'train']
{'adidas': 0, 'nike': 1}
Shape of X [N, C, H, W]:  torch.Size([32, 3, 224, 224])
Shape of y:  torch.Size([32]) torch.int64
Using cpu device
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)
  )
)
Epoch: 1, Train_acc:54.8%, Train_loss:0.743, Test_acc:57.9%, Test_loss:0.681, Lr:1.00E-04
Epoch: 2, Train_acc:62.2%, Train_loss:0.666, Test_acc:67.1%, Test_loss:0.641, Lr:9.20E-05
Epoch: 3, Train_acc:67.1%, Train_loss:0.608, Test_acc:64.5%, Test_loss:0.583, Lr:9.20E-05
Epoch: 4, Train_acc:73.9%, Train_loss:0.552, Test_acc:65.8%, Test_loss:0.569, Lr:8.46E-05
Epoch: 5, Train_acc:73.9%, Train_loss:0.544, Test_acc:73.7%, Test_loss:0.535, Lr:8.46E-05
Epoch: 6, Train_acc:80.3%, Train_loss:0.484, Test_acc:75.0%, Test_loss:0.530, Lr:7.79E-05
Epoch: 7, Train_acc:80.7%, Train_loss:0.450, Test_acc:76.3%, Test_loss:0.538, Lr:7.79E-05
Epoch: 8, Train_acc:83.1%, Train_loss:0.446, Test_acc:72.4%, Test_loss:0.515, Lr:7.16E-05
Epoch: 9, Train_acc:82.9%, Train_loss:0.427, Test_acc:78.9%, Test_loss:0.487, Lr:7.16E-05
Epoch:10, Train_acc:84.1%, Train_loss:0.406, Test_acc:76.3%, Test_loss:0.510, Lr:6.59E-05
Epoch:11, Train_acc:86.9%, Train_loss:0.393, Test_acc:76.3%, Test_loss:0.516, Lr:6.59E-05
Epoch:12, Train_acc:89.4%, Train_loss:0.375, Test_acc:77.6%, Test_loss:0.475, Lr:6.06E-05
Epoch:13, Train_acc:86.9%, Train_loss:0.377, Test_acc:77.6%, Test_loss:0.468, Lr:6.06E-05
Epoch:14, Train_acc:91.0%, Train_loss:0.350, Test_acc:80.3%, Test_loss:0.457, Lr:5.58E-05
Epoch:15, Train_acc:91.0%, Train_loss:0.335, Test_acc:78.9%, Test_loss:0.452, Lr:5.58E-05
Epoch:16, Train_acc:91.4%, Train_loss:0.331, Test_acc:77.6%, Test_loss:0.468, Lr:5.13E-05
Epoch:17, Train_acc:89.2%, Train_loss:0.335, Test_acc:76.3%, Test_loss:0.465, Lr:5.13E-05
Epoch:18, Train_acc:91.8%, Train_loss:0.321, Test_acc:77.6%, Test_loss:0.442, Lr:4.72E-05
Epoch:19, Train_acc:91.0%, Train_loss:0.309, Test_acc:77.6%, Test_loss:0.493, Lr:4.72E-05
Epoch:20, Train_acc:91.8%, Train_loss:0.311, Test_acc:78.9%, Test_loss:0.450, Lr:4.34E-05
Epoch:21, Train_acc:93.2%, Train_loss:0.300, Test_acc:77.6%, Test_loss:0.455, Lr:4.34E-05
Epoch:22, Train_acc:94.0%, Train_loss:0.283, Test_acc:77.6%, Test_loss:0.420, Lr:4.00E-05
Epoch:23, Train_acc:94.4%, Train_loss:0.282, Test_acc:78.9%, Test_loss:0.514, Lr:4.00E-05
Epoch:24, Train_acc:92.4%, Train_loss:0.301, Test_acc:77.6%, Test_loss:0.452, Lr:3.68E-05
Epoch:25, Train_acc:94.4%, Train_loss:0.275, Test_acc:77.6%, Test_loss:0.424, Lr:3.68E-05
Epoch:26, Train_acc:94.2%, Train_loss:0.266, Test_acc:77.6%, Test_loss:0.423, Lr:3.38E-05
Epoch:27, Train_acc:95.8%, Train_loss:0.273, Test_acc:77.6%, Test_loss:0.421, Lr:3.38E-05
Epoch:28, Train_acc:94.4%, Train_loss:0.275, Test_acc:77.6%, Test_loss:0.407, Lr:3.11E-05
Epoch:29, Train_acc:96.2%, Train_loss:0.264, Test_acc:77.6%, Test_loss:0.499, Lr:3.11E-05
Epoch:30, Train_acc:95.4%, Train_loss:0.263, Test_acc:77.6%, Test_loss:0.488, Lr:2.86E-05
Epoch:31, Train_acc:95.0%, Train_loss:0.256, Test_acc:77.6%, Test_loss:0.398, Lr:2.86E-05
Epoch:32, Train_acc:96.2%, Train_loss:0.263, Test_acc:77.6%, Test_loss:0.418, Lr:2.63E-05
Epoch:33, Train_acc:96.0%, Train_loss:0.258, Test_acc:77.6%, Test_loss:0.416, Lr:2.63E-05
Epoch:34, Train_acc:94.0%, Train_loss:0.260, Test_acc:77.6%, Test_loss:0.423, Lr:2.42E-05
Epoch:35, Train_acc:95.8%, Train_loss:0.250, Test_acc:77.6%, Test_loss:0.419, Lr:2.42E-05
Epoch:36, Train_acc:95.8%, Train_loss:0.251, Test_acc:78.9%, Test_loss:0.404, Lr:2.23E-05
Epoch:37, Train_acc:95.6%, Train_loss:0.243, Test_acc:77.6%, Test_loss:0.443, Lr:2.23E-05
Epoch:38, Train_acc:95.8%, Train_loss:0.247, Test_acc:77.6%, Test_loss:0.420, Lr:2.05E-05
Epoch:39, Train_acc:96.8%, Train_loss:0.232, Test_acc:77.6%, Test_loss:0.421, Lr:2.05E-05
Epoch:40, Train_acc:97.2%, Train_loss:0.242, Test_acc:78.9%, Test_loss:0.395, Lr:1.89E-05
Done
预测结果是：adidas

进程已结束，退出代码为 0

调用官方动态学习率接口-StepLR

D:\Programs\Python\Python39\python.exe D:\PycharmProjects\pythonProject\P5\main.py 
cpu
['test', 'train']
{'adidas': 0, 'nike': 1}
Shape of X [N, C, H, W]:  torch.Size([32, 3, 224, 224])
Shape of y:  torch.Size([32]) torch.int64
Using cpu device
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)
  )
)
Epoch: 1, Train_acc:53.0%, Train_loss:2.723, Test_acc:50.0%, Test_loss:1.418, Lr:1.00E-03
Epoch: 2, Train_acc:53.2%, Train_loss:2.109, Test_acc:50.0%, Test_loss:1.946, Lr:1.00E-03
Epoch: 3, Train_acc:61.0%, Train_loss:1.290, Test_acc:65.8%, Test_loss:1.072, Lr:1.00E-03
Epoch: 4, Train_acc:65.1%, Train_loss:1.089, Test_acc:57.9%, Test_loss:1.579, Lr:1.00E-03
Epoch: 5, Train_acc:62.9%, Train_loss:1.048, Test_acc:69.7%, Test_loss:0.602, Lr:1.00E-04
Epoch: 6, Train_acc:85.9%, Train_loss:0.313, Test_acc:68.4%, Test_loss:0.509, Lr:1.00E-04
Epoch: 7, Train_acc:86.9%, Train_loss:0.311, Test_acc:68.4%, Test_loss:0.634, Lr:1.00E-04
Epoch: 8, Train_acc:89.4%, Train_loss:0.281, Test_acc:71.1%, Test_loss:0.516, Lr:1.00E-04
Epoch: 9, Train_acc:89.2%, Train_loss:0.282, Test_acc:67.1%, Test_loss:0.545, Lr:1.00E-04
Epoch:10, Train_acc:89.2%, Train_loss:0.283, Test_acc:69.7%, Test_loss:0.554, Lr:1.00E-05
Epoch:11, Train_acc:90.0%, Train_loss:0.272, Test_acc:69.7%, Test_loss:0.594, Lr:1.00E-05
Epoch:12, Train_acc:89.6%, Train_loss:0.285, Test_acc:69.7%, Test_loss:0.497, Lr:1.00E-05
Epoch:13, Train_acc:89.2%, Train_loss:0.274, Test_acc:69.7%, Test_loss:0.491, Lr:1.00E-05
Epoch:14, Train_acc:88.6%, Train_loss:0.267, Test_acc:69.7%, Test_loss:0.552, Lr:1.00E-05
Epoch:15, Train_acc:89.4%, Train_loss:0.269, Test_acc:69.7%, Test_loss:0.531, Lr:1.00E-06
Epoch:16, Train_acc:88.6%, Train_loss:0.269, Test_acc:69.7%, Test_loss:0.549, Lr:1.00E-06
Epoch:17, Train_acc:89.4%, Train_loss:0.271, Test_acc:69.7%, Test_loss:0.513, Lr:1.00E-06
Epoch:18, Train_acc:90.2%, Train_loss:0.272, Test_acc:69.7%, Test_loss:0.524, Lr:1.00E-06
Epoch:19, Train_acc:89.2%, Train_loss:0.262, Test_acc:69.7%, Test_loss:0.532, Lr:1.00E-06
Epoch:20, Train_acc:90.0%, Train_loss:0.262, Test_acc:69.7%, Test_loss:0.634, Lr:1.00E-07
Epoch:21, Train_acc:90.4%, Train_loss:0.265, Test_acc:69.7%, Test_loss:0.549, Lr:1.00E-07
Epoch:22, Train_acc:89.2%, Train_loss:0.276, Test_acc:69.7%, Test_loss:0.547, Lr:1.00E-07
Epoch:23, Train_acc:90.8%, Train_loss:0.269, Test_acc:69.7%, Test_loss:0.585, Lr:1.00E-07
Epoch:24, Train_acc:90.4%, Train_loss:0.269, Test_acc:69.7%, Test_loss:0.491, Lr:1.00E-07
Epoch:25, Train_acc:89.0%, Train_loss:0.271, Test_acc:69.7%, Test_loss:0.502, Lr:1.00E-08
Epoch:26, Train_acc:89.0%, Train_loss:0.279, Test_acc:69.7%, Test_loss:0.524, Lr:1.00E-08
Epoch:27, Train_acc:89.4%, Train_loss:0.278, Test_acc:69.7%, Test_loss:0.523, Lr:1.00E-08
Epoch:28, Train_acc:88.8%, Train_loss:0.266, Test_acc:69.7%, Test_loss:0.524, Lr:1.00E-08
Epoch:29, Train_acc:89.8%, Train_loss:0.275, Test_acc:69.7%, Test_loss:0.480, Lr:1.00E-08
Epoch:30, Train_acc:89.0%, Train_loss:0.276, Test_acc:69.7%, Test_loss:0.525, Lr:1.00E-09
Epoch:31, Train_acc:89.6%, Train_loss:0.271, Test_acc:69.7%, Test_loss:0.548, Lr:1.00E-09
Epoch:32, Train_acc:89.8%, Train_loss:0.268, Test_acc:69.7%, Test_loss:0.549, Lr:1.00E-09
Epoch:33, Train_acc:88.4%, Train_loss:0.282, Test_acc:69.7%, Test_loss:0.473, Lr:1.00E-09
Epoch:34, Train_acc:90.4%, Train_loss:0.265, Test_acc:69.7%, Test_loss:0.514, Lr:1.00E-09
Epoch:35, Train_acc:89.2%, Train_loss:0.270, Test_acc:69.7%, Test_loss:0.514, Lr:1.00E-10
Epoch:36, Train_acc:89.8%, Train_loss:0.266, Test_acc:69.7%, Test_loss:0.496, Lr:1.00E-10
Epoch:37, Train_acc:90.0%, Train_loss:0.274, Test_acc:69.7%, Test_loss:0.545, Lr:1.00E-10
Epoch:38, Train_acc:90.6%, Train_loss:0.258, Test_acc:69.7%, Test_loss:0.577, Lr:1.00E-10
Epoch:39, Train_acc:90.4%, Train_loss:0.269, Test_acc:69.7%, Test_loss:0.514, Lr:1.00E-10
Epoch:40, Train_acc:88.4%, Train_loss:0.272, Test_acc:69.7%, Test_loss:0.522, Lr:1.00E-11
Done
预测结果是：adidas

进程已结束，退出代码为 0

调用官方动态学习率接口-MultiStepLR

D:\Programs\Python\Python39\python.exe D:\PycharmProjects\pythonProject\P5\main.py 
cpu
['test', 'train']
{'adidas': 0, 'nike': 1}
Shape of X [N, C, H, W]:  torch.Size([32, 3, 224, 224])
Shape of y:  torch.Size([32]) torch.int64
Using cpu device
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)
  )
)
Epoch: 1, Train_acc:49.0%, Train_loss:3.841, Test_acc:50.0%, Test_loss:1.384, Lr:1.00E-03
Epoch: 2, Train_acc:58.6%, Train_loss:1.675, Test_acc:57.9%, Test_loss:1.427, Lr:1.00E-04
Epoch: 3, Train_acc:74.5%, Train_loss:0.626, Test_acc:71.1%, Test_loss:0.541, Lr:1.00E-04
Epoch: 4, Train_acc:80.1%, Train_loss:0.420, Test_acc:71.1%, Test_loss:0.567, Lr:1.00E-04
Epoch: 5, Train_acc:83.5%, Train_loss:0.382, Test_acc:72.4%, Test_loss:0.513, Lr:1.00E-04
Epoch: 6, Train_acc:83.5%, Train_loss:0.361, Test_acc:72.4%, Test_loss:0.511, Lr:1.00E-05
Epoch: 7, Train_acc:85.5%, Train_loss:0.358, Test_acc:71.1%, Test_loss:0.545, Lr:1.00E-05
Epoch: 8, Train_acc:86.9%, Train_loss:0.332, Test_acc:76.3%, Test_loss:0.548, Lr:1.00E-05
Epoch: 9, Train_acc:85.3%, Train_loss:0.346, Test_acc:75.0%, Test_loss:0.502, Lr:1.00E-05
Epoch:10, Train_acc:85.9%, Train_loss:0.338, Test_acc:72.4%, Test_loss:0.525, Lr:1.00E-05
Epoch:11, Train_acc:83.5%, Train_loss:0.345, Test_acc:72.4%, Test_loss:0.508, Lr:1.00E-05
Epoch:12, Train_acc:85.1%, Train_loss:0.335, Test_acc:73.7%, Test_loss:0.472, Lr:1.00E-05
Epoch:13, Train_acc:85.3%, Train_loss:0.328, Test_acc:72.4%, Test_loss:0.483, Lr:1.00E-05
Epoch:14, Train_acc:86.1%, Train_loss:0.324, Test_acc:72.4%, Test_loss:0.485, Lr:1.00E-05
Epoch:15, Train_acc:85.9%, Train_loss:0.329, Test_acc:72.4%, Test_loss:0.522, Lr:1.00E-06
Epoch:16, Train_acc:86.7%, Train_loss:0.326, Test_acc:72.4%, Test_loss:0.496, Lr:1.00E-06
Epoch:17, Train_acc:84.5%, Train_loss:0.331, Test_acc:72.4%, Test_loss:0.536, Lr:1.00E-06
Epoch:18, Train_acc:86.5%, Train_loss:0.323, Test_acc:72.4%, Test_loss:0.538, Lr:1.00E-06
Epoch:19, Train_acc:85.9%, Train_loss:0.320, Test_acc:72.4%, Test_loss:0.504, Lr:1.00E-06
Epoch:20, Train_acc:85.5%, Train_loss:0.331, Test_acc:72.4%, Test_loss:0.499, Lr:1.00E-06
Epoch:21, Train_acc:86.5%, Train_loss:0.325, Test_acc:72.4%, Test_loss:0.481, Lr:1.00E-06
Epoch:22, Train_acc:85.7%, Train_loss:0.329, Test_acc:72.4%, Test_loss:0.479, Lr:1.00E-06
Epoch:23, Train_acc:85.7%, Train_loss:0.326, Test_acc:72.4%, Test_loss:0.449, Lr:1.00E-06
Epoch:24, Train_acc:86.5%, Train_loss:0.326, Test_acc:72.4%, Test_loss:0.548, Lr:1.00E-06
Epoch:25, Train_acc:85.3%, Train_loss:0.330, Test_acc:72.4%, Test_loss:0.531, Lr:1.00E-06
Epoch:26, Train_acc:85.3%, Train_loss:0.327, Test_acc:72.4%, Test_loss:0.455, Lr:1.00E-06
Epoch:27, Train_acc:86.3%, Train_loss:0.328, Test_acc:72.4%, Test_loss:0.503, Lr:1.00E-06
Epoch:28, Train_acc:87.1%, Train_loss:0.325, Test_acc:72.4%, Test_loss:0.496, Lr:1.00E-06
Epoch:29, Train_acc:86.1%, Train_loss:0.324, Test_acc:72.4%, Test_loss:0.515, Lr:1.00E-06
Epoch:30, Train_acc:86.5%, Train_loss:0.320, Test_acc:73.7%, Test_loss:0.463, Lr:1.00E-06
Epoch:31, Train_acc:85.9%, Train_loss:0.332, Test_acc:72.4%, Test_loss:0.496, Lr:1.00E-06
Epoch:32, Train_acc:85.1%, Train_loss:0.339, Test_acc:72.4%, Test_loss:0.522, Lr:1.00E-06
Epoch:33, Train_acc:86.3%, Train_loss:0.326, Test_acc:72.4%, Test_loss:0.517, Lr:1.00E-06
Epoch:34, Train_acc:86.7%, Train_loss:0.327, Test_acc:72.4%, Test_loss:0.509, Lr:1.00E-06
Epoch:35, Train_acc:85.7%, Train_loss:0.326, Test_acc:72.4%, Test_loss:0.530, Lr:1.00E-06
Epoch:36, Train_acc:85.5%, Train_loss:0.323, Test_acc:72.4%, Test_loss:0.555, Lr:1.00E-06
Epoch:37, Train_acc:85.7%, Train_loss:0.314, Test_acc:72.4%, Test_loss:0.473, Lr:1.00E-06
Epoch:38, Train_acc:84.5%, Train_loss:0.337, Test_acc:72.4%, Test_loss:0.497, Lr:1.00E-06
Epoch:39, Train_acc:85.5%, Train_loss:0.339, Test_acc:72.4%, Test_loss:0.448, Lr:1.00E-06
Epoch:40, Train_acc:86.3%, Train_loss:0.315, Test_acc:72.4%, Test_loss:0.494, Lr:1.00E-06
Done
预测结果是：adidas

进程已结束，退出代码为 0

二、源代码

#一、 前期准备
#1. 设置GPU

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

import os,PIL,pathlib

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

print(device)

#2. 导入数据
import os,PIL,random,pathlib

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

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

# 关于transforms.Compose的更多介绍可以参考：https://blog.youkuaiyun.com/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。
])

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("./data/train/",transform=train_transforms)
test_dataset  = datasets.ImageFolder("./data/test/",transform=test_transform)

print(train_dataset.class_to_idx)

batch_size = 32

train_dl = torch.utils.data.DataLoader(train_dataset,
                                           batch_size=batch_size,
                                           shuffle=True, #如果设置为 True，则在每个 epoch 开始时对数据进行洗牌，以随机打乱样本的顺序。这对于训练数据的随机性很重要，以避免模型学习到数据的顺序性。默认值为 False
                                           num_workers=0) #用于数据加载的子进程数量。通常，将其设置为大于 0 的值可以加快数据加载速度，特别是当数据集很大时。默认值为 0，表示在主进程中加载数据。
test_dl = torch.utils.data.DataLoader(test_dataset,
                                          batch_size=batch_size,
                                          shuffle=True,
                                          num_workers=0)

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

#二、构建简单的CNN网络

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(classeNames)))

    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


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

model = Model().to(device)
print(model)

#三、 训练模型
#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
    num_batches = len(dataloader)  # 批次数目, (size/batch_size，向上取整)

    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

#2. 编写测试函数
def test(dataloader, model, loss_fn):
    size = len(dataloader.dataset)  # 测试集的大小
    num_batches = len(dataloader)  # 批次数目, (size/batch_size，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. 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 0.92
#     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)


# # 调用官方动态学习率接口时使用LambdaLR
learn_rate = 1e-4 # 初始学习率
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


# # 调用官方动态学习率接口时使用StepLR
# optimizer = torch.optim.SGD(model.parameters(), lr=0.001 )
# scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=5, gamma=0.1)

# # 调用官方动态学习率接口时使用MultiStepLR
# optimizer = torch.optim.SGD(model.parameters(), lr=0.001 )
# scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer,
#                                                  milestones=[2,6,15], #调整学习率的epoch数
#                                                  gamma=0.1)

#4. 正式训练
loss_fn = nn.CrossEntropyLoss()  # 创建损失函数
epochs = 40

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

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)
    scheduler.step() # 更新学习率（调用官方动态学习率接口时使用）

    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))
print('Done')

#四、 结果可视化
#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

from datetime import datetime
current_time = datetime.now() # 获取当前时间

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.xlabel(current_time) # 打卡请带上时间戳，否则代码截图无效

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()

#2. 指定图片进行预测

from PIL import Image

classes = list(train_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.to(device).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}')

# 预测训练集中的某张照片
predict_one_image(image_path='./data/test/adidas/1.jpg',
                  model=model,
                  transform=train_transforms,
                  classes=classes)


#五、保存并加载模型
# 模型保存
PATH = './model.pth'  # 保存的参数文件名
torch.save(model.state_dict(), PATH)

# 将参数加载到model当中
model.load_state_dict(torch.load(PATH, map_location=device))

您可能感兴趣的与本文相关的镜像

PyTorch 2.5

PyTorch

Cuda

PyTorch 是一个开源的 Python 机器学习库，基于 Torch 库，底层由 C++ 实现，应用于人工智能领域，如计算机视觉和自然语言处理