深度学习识别CIFAR10：pytorch训练LeNet、AlexNet、VGG19实现及比较（二）

2023-07-24 22:31:52

版权声明：本文为博主原创文章，欢迎转载，并请注明出处。联系方式：460356155@qq.com

AlexNet在2012年ImageNet图像分类任务竞赛中获得冠军。网络结构如下图所示：

深度学习识别CIFAR10：pytorch训练LeNet、AlexNet、VGG19实现及比较（二）

对CIFAR10，图片是32*32，尺寸远小于227*227，因此对网络结构和参数需做微调：

卷积层1：核大小7*7，步长2，填充2

最后一个max-pool层删除

网络定义代码如下：

 class AlexNet(nn.Module):

     def __init__(self):

         super(AlexNet, self).__init__()

         self.cnn = nn.Sequential(

             # 卷积层1，3通道输入，96个卷积核，核大小7*7，步长2，填充2

             # 经过该层图像大小变为32-7+2*2 / 2 +1，15*15

             # 经3*3最大池化，2步长，图像变为15-3 / 2 + 1， 7*7

             nn.Conv2d(3, 96, 7, 2, 2),

             nn.ReLU(inplace=True),

             nn.MaxPool2d(3, 2, 0),

             # 卷积层2，96输入通道，256个卷积核，核大小5*5，步长1，填充2

             # 经过该层图像变为7-5+2*2 / 1 + 1，7*7

             # 经3*3最大池化，2步长，图像变为7-3 / 2 + 1， 3*3

             nn.Conv2d(96, 256, 5, 1, 2),

             nn.ReLU(inplace=True),

             nn.MaxPool2d(3, 2, 0),

             # 卷积层3，256输入通道，384个卷积核，核大小3*3，步长1，填充1

             # 经过该层图像变为3-3+2*1 / 1 + 1，3*3

             nn.Conv2d(256, 384, 3, 1, 1),

             nn.ReLU(inplace=True),

             # 卷积层3，384输入通道，384个卷积核，核大小3*3，步长1，填充1

             # 经过该层图像变为3-3+2*1 / 1 + 1，3*3

             nn.Conv2d(384, 384, 3, 1, 1),

             nn.ReLU(inplace=True),

             # 卷积层3，384输入通道，256个卷积核，核大小3*3，步长1，填充1

             # 经过该层图像变为3-3+2*1 / 1 + 1，3*3

             nn.Conv2d(384, 256, 3, 1, 1),

             nn.ReLU(inplace=True)

         )

         self.fc = nn.Sequential(

             # 256个feature，每个feature 3*3

             nn.Linear(256*3*3, 1024),

             nn.ReLU(),

             nn.Linear(1024, 512),

             nn.ReLU(),

             nn.Linear(512, 10)

         )

     def forward(self, x):

         x = self.cnn(x)

         # x.size()[0]: batch size

         x = x.view(x.size()[0], -1)

         x = self.fc(x)

         return x

其余代码同深度学习识别CIFAR10：pytorch训练LeNet、AlexNet、VGG19实现及比较（一）。运行结果如下：

Files already downloaded and verified
AlexNet(
  (cnn): Sequential(
    (0): Conv2d(3, 96, kernel_size=(7, 7), stride=(2, 2), padding=(2, 2))
    (1): ReLU(inplace)
    (2): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
    (3): Conv2d(96, 256, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
    (4): ReLU(inplace)
    (5): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
    (6): Conv2d(256, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (7): ReLU(inplace)
    (8): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (9): ReLU(inplace)
    (10): Conv2d(384, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (11): ReLU(inplace)
  )
  (fc): Sequential(
    (0): Linear(in_features=2304, out_features=1024, bias=True)
    (1): ReLU()
    (2): Linear(in_features=1024, out_features=512, bias=True)
    (3): ReLU()
    (4): Linear(in_features=512, out_features=10, bias=True)
  )
)
Train Epoch: 1 [6400/50000 (13%)]    Loss: 2.303003  Acc: 10.000000
Train Epoch: 1 [12800/50000 (26%)]    Loss: 2.302847  Acc: 9.000000
Train Epoch: 1 [19200/50000 (38%)]    Loss: 2.302748  Acc: 9.000000
Train Epoch: 1 [25600/50000 (51%)]    Loss: 2.302349  Acc: 10.000000
Train Epoch: 1 [32000/50000 (64%)]    Loss: 2.301069  Acc: 10.000000
Train Epoch: 1 [38400/50000 (77%)]    Loss: 2.275476  Acc: 12.000000
Train Epoch: 1 [44800/50000 (90%)]    Loss: 2.231073  Acc: 13.000000
one epoch spend:  0:00:06.866484
EPOCH:1, ACC:25.06

Train Epoch: 2 [6400/50000 (13%)]    Loss: 1.848806  Acc: 25.000000
Train Epoch: 2 [12800/50000 (26%)]    Loss: 1.808251  Acc: 27.000000
Train Epoch: 2 [19200/50000 (38%)]    Loss: 1.774210  Acc: 29.000000
Train Epoch: 2 [25600/50000 (51%)]    Loss: 1.744809  Acc: 31.000000
Train Epoch: 2 [32000/50000 (64%)]    Loss: 1.714098  Acc: 32.000000
Train Epoch: 2 [38400/50000 (77%)]    Loss: 1.684451  Acc: 34.000000
Train Epoch: 2 [44800/50000 (90%)]    Loss: 1.654931  Acc: 35.000000
one epoch spend:  0:00:06.941943
EPOCH:2, ACC:46.64

Train Epoch: 3 [6400/50000 (13%)]    Loss: 1.418345  Acc: 45.000000
Train Epoch: 3 [12800/50000 (26%)]    Loss: 1.368839  Acc: 47.000000
Train Epoch: 3 [19200/50000 (38%)]    Loss: 1.349170  Acc: 48.000000
Train Epoch: 3 [25600/50000 (51%)]    Loss: 1.326504  Acc: 49.000000
Train Epoch: 3 [32000/50000 (64%)]    Loss: 1.316630  Acc: 50.000000
Train Epoch: 3 [38400/50000 (77%)]    Loss: 1.300982  Acc: 51.000000
Train Epoch: 3 [44800/50000 (90%)]    Loss: 1.288368  Acc: 52.000000
one epoch spend:  0:00:07.031582
EPOCH:3, ACC:56.72

Train Epoch: 4 [6400/50000 (13%)]    Loss: 1.078210  Acc: 60.000000
Train Epoch: 4 [12800/50000 (26%)]    Loss: 1.083730  Acc: 60.000000
Train Epoch: 4 [19200/50000 (38%)]    Loss: 1.085976  Acc: 60.000000
Train Epoch: 4 [25600/50000 (51%)]    Loss: 1.080863  Acc: 61.000000
Train Epoch: 4 [32000/50000 (64%)]    Loss: 1.076230  Acc: 61.000000
Train Epoch: 4 [38400/50000 (77%)]    Loss: 1.067998  Acc: 61.000000
Train Epoch: 4 [44800/50000 (90%)]    Loss: 1.058093  Acc: 62.000000
one epoch spend:  0:00:06.908232
EPOCH:4, ACC:65.4

Train Epoch: 5 [6400/50000 (13%)]    Loss: 0.911678  Acc: 67.000000
Train Epoch: 5 [12800/50000 (26%)]    Loss: 0.904799  Acc: 67.000000
Train Epoch: 5 [19200/50000 (38%)]    Loss: 0.914306  Acc: 67.000000
Train Epoch: 5 [25600/50000 (51%)]    Loss: 0.906587  Acc: 67.000000
Train Epoch: 5 [32000/50000 (64%)]    Loss: 0.902747  Acc: 67.000000
Train Epoch: 5 [38400/50000 (77%)]    Loss: 0.896548  Acc: 68.000000
Train Epoch: 5 [44800/50000 (90%)]    Loss: 0.895071  Acc: 68.000000
one epoch spend:  0:00:06.868743
EPOCH:5, ACC:66.47

Train Epoch: 6 [6400/50000 (13%)]    Loss: 0.769778  Acc: 72.000000
Train Epoch: 6 [12800/50000 (26%)]    Loss: 0.770126  Acc: 73.000000
Train Epoch: 6 [19200/50000 (38%)]    Loss: 0.775755  Acc: 72.000000
Train Epoch: 6 [25600/50000 (51%)]    Loss: 0.775044  Acc: 72.000000
Train Epoch: 6 [32000/50000 (64%)]    Loss: 0.772686  Acc: 72.000000
Train Epoch: 6 [38400/50000 (77%)]    Loss: 0.765352  Acc: 73.000000
Train Epoch: 6 [44800/50000 (90%)]    Loss: 0.768808  Acc: 73.000000
one epoch spend:  0:00:06.868047
EPOCH:6, ACC:68.26

Train Epoch: 7 [6400/50000 (13%)]    Loss: 0.641943  Acc: 77.000000
Train Epoch: 7 [12800/50000 (26%)]    Loss: 0.643955  Acc: 77.000000
Train Epoch: 7 [19200/50000 (38%)]    Loss: 0.642063  Acc: 77.000000
Train Epoch: 7 [25600/50000 (51%)]    Loss: 0.647976  Acc: 77.000000
Train Epoch: 7 [32000/50000 (64%)]    Loss: 0.648042  Acc: 77.000000
Train Epoch: 7 [38400/50000 (77%)]    Loss: 0.652435  Acc: 77.000000
Train Epoch: 7 [44800/50000 (90%)]    Loss: 0.655997  Acc: 77.000000
one epoch spend:  0:00:06.962986
EPOCH:7, ACC:72.21

Train Epoch: 8 [6400/50000 (13%)]    Loss: 0.541914  Acc: 80.000000
Train Epoch: 8 [12800/50000 (26%)]    Loss: 0.543631  Acc: 81.000000
Train Epoch: 8 [19200/50000 (38%)]    Loss: 0.551045  Acc: 80.000000
Train Epoch: 8 [25600/50000 (51%)]    Loss: 0.551447  Acc: 80.000000
Train Epoch: 8 [32000/50000 (64%)]    Loss: 0.554876  Acc: 80.000000
Train Epoch: 8 [38400/50000 (77%)]    Loss: 0.560712  Acc: 80.000000
Train Epoch: 8 [44800/50000 (90%)]    Loss: 0.561110  Acc: 80.000000
one epoch spend:  0:00:07.025618
EPOCH:8, ACC:74.15

Train Epoch: 9 [6400/50000 (13%)]    Loss: 0.452407  Acc: 84.000000
Train Epoch: 9 [12800/50000 (26%)]    Loss: 0.462235  Acc: 83.000000
Train Epoch: 9 [19200/50000 (38%)]    Loss: 0.476642  Acc: 83.000000
Train Epoch: 9 [25600/50000 (51%)]    Loss: 0.478906  Acc: 83.000000
Train Epoch: 9 [32000/50000 (64%)]    Loss: 0.476015  Acc: 83.000000
Train Epoch: 9 [38400/50000 (77%)]    Loss: 0.477935  Acc: 83.000000
Train Epoch: 9 [44800/50000 (90%)]    Loss: 0.480251  Acc: 83.000000
one epoch spend:  0:00:06.840690
EPOCH:9, ACC:74.49

Train Epoch: 10 [6400/50000 (13%)]    Loss: 0.383466  Acc: 87.000000
Train Epoch: 10 [12800/50000 (26%)]    Loss: 0.376466  Acc: 87.000000
Train Epoch: 10 [19200/50000 (38%)]    Loss: 0.386534  Acc: 86.000000
Train Epoch: 10 [25600/50000 (51%)]    Loss: 0.394657  Acc: 86.000000
Train Epoch: 10 [32000/50000 (64%)]    Loss: 0.394315  Acc: 86.000000
Train Epoch: 10 [38400/50000 (77%)]    Loss: 0.395472  Acc: 86.000000
Train Epoch: 10 [44800/50000 (90%)]    Loss: 0.399573  Acc: 86.000000
one epoch spend:  0:00:06.866040
EPOCH:10, ACC:73.13

Train Epoch: 11 [6400/50000 (13%)]    Loss: 0.297959  Acc: 89.000000
Train Epoch: 11 [12800/50000 (26%)]    Loss: 0.305871  Acc: 89.000000
Train Epoch: 11 [19200/50000 (38%)]    Loss: 0.315880  Acc: 89.000000
Train Epoch: 11 [25600/50000 (51%)]    Loss: 0.322634  Acc: 88.000000
Train Epoch: 11 [32000/50000 (64%)]    Loss: 0.326418  Acc: 88.000000
Train Epoch: 11 [38400/50000 (77%)]    Loss: 0.333330  Acc: 88.000000
Train Epoch: 11 [44800/50000 (90%)]    Loss: 0.337955  Acc: 88.000000
one epoch spend:  0:00:06.884786
EPOCH:11, ACC:73.79

Train Epoch: 12 [6400/50000 (13%)]    Loss: 0.242202  Acc: 91.000000
Train Epoch: 12 [12800/50000 (26%)]    Loss: 0.250616  Acc: 91.000000
Train Epoch: 12 [19200/50000 (38%)]    Loss: 0.265347  Acc: 90.000000
Train Epoch: 12 [25600/50000 (51%)]    Loss: 0.271456  Acc: 90.000000
Train Epoch: 12 [32000/50000 (64%)]    Loss: 0.273988  Acc: 90.000000
Train Epoch: 12 [38400/50000 (77%)]    Loss: 0.280836  Acc: 90.000000
Train Epoch: 12 [44800/50000 (90%)]    Loss: 0.281419  Acc: 90.000000
one epoch spend:  0:00:06.906915
EPOCH:12, ACC:75.89

Train Epoch: 13 [6400/50000 (13%)]    Loss: 0.228122  Acc: 92.000000
Train Epoch: 13 [12800/50000 (26%)]    Loss: 0.228350  Acc: 92.000000
Train Epoch: 13 [19200/50000 (38%)]    Loss: 0.227151  Acc: 92.000000
Train Epoch: 13 [25600/50000 (51%)]    Loss: 0.228918  Acc: 92.000000
Train Epoch: 13 [32000/50000 (64%)]    Loss: 0.232642  Acc: 91.000000
Train Epoch: 13 [38400/50000 (77%)]    Loss: 0.237782  Acc: 91.000000
Train Epoch: 13 [44800/50000 (90%)]    Loss: 0.242339  Acc: 91.000000
one epoch spend:  0:00:06.869576
EPOCH:13, ACC:74.39

Train Epoch: 14 [6400/50000 (13%)]    Loss: 0.179683  Acc: 93.000000
Train Epoch: 14 [12800/50000 (26%)]    Loss: 0.182840  Acc: 93.000000
Train Epoch: 14 [19200/50000 (38%)]    Loss: 0.182861  Acc: 93.000000
Train Epoch: 14 [25600/50000 (51%)]    Loss: 0.189549  Acc: 93.000000
Train Epoch: 14 [32000/50000 (64%)]    Loss: 0.193639  Acc: 93.000000
Train Epoch: 14 [38400/50000 (77%)]    Loss: 0.196073  Acc: 93.000000
Train Epoch: 14 [44800/50000 (90%)]    Loss: 0.198425  Acc: 93.000000
one epoch spend:  0:00:06.927269
EPOCH:14, ACC:75.63

Train Epoch: 15 [6400/50000 (13%)]    Loss: 0.123262  Acc: 95.000000
Train Epoch: 15 [12800/50000 (26%)]    Loss: 0.136458  Acc: 95.000000
Train Epoch: 15 [19200/50000 (38%)]    Loss: 0.141503  Acc: 95.000000
Train Epoch: 15 [25600/50000 (51%)]    Loss: 0.147542  Acc: 94.000000
Train Epoch: 15 [32000/50000 (64%)]    Loss: 0.149795  Acc: 94.000000
Train Epoch: 15 [38400/50000 (77%)]    Loss: 0.154987  Acc: 94.000000
Train Epoch: 15 [44800/50000 (90%)]    Loss: 0.157952  Acc: 94.000000
one epoch spend:  0:00:07.015382
EPOCH:15, ACC:74.6

Train Epoch: 16 [6400/50000 (13%)]    Loss: 0.144001  Acc: 94.000000
Train Epoch: 16 [12800/50000 (26%)]    Loss: 0.141813  Acc: 94.000000
Train Epoch: 16 [19200/50000 (38%)]    Loss: 0.139413  Acc: 95.000000
Train Epoch: 16 [25600/50000 (51%)]    Loss: 0.136546  Acc: 95.000000
Train Epoch: 16 [32000/50000 (64%)]    Loss: 0.138039  Acc: 95.000000
Train Epoch: 16 [38400/50000 (77%)]    Loss: 0.139393  Acc: 95.000000
Train Epoch: 16 [44800/50000 (90%)]    Loss: 0.142776  Acc: 95.000000
one epoch spend:  0:00:06.883968
EPOCH:16, ACC:75.54

Train Epoch: 17 [6400/50000 (13%)]    Loss: 0.080704  Acc: 97.000000
Train Epoch: 17 [12800/50000 (26%)]    Loss: 0.098754  Acc: 96.000000
Train Epoch: 17 [19200/50000 (38%)]    Loss: 0.104385  Acc: 96.000000
Train Epoch: 17 [25600/50000 (51%)]    Loss: 0.107634  Acc: 96.000000
Train Epoch: 17 [32000/50000 (64%)]    Loss: 0.112148  Acc: 96.000000
Train Epoch: 17 [38400/50000 (77%)]    Loss: 0.113687  Acc: 96.000000
Train Epoch: 17 [44800/50000 (90%)]    Loss: 0.114508  Acc: 96.000000
one epoch spend:  0:00:06.905244
EPOCH:17, ACC:74.9

Train Epoch: 18 [6400/50000 (13%)]    Loss: 0.085284  Acc: 97.000000
Train Epoch: 18 [12800/50000 (26%)]    Loss: 0.087985  Acc: 97.000000
Train Epoch: 18 [19200/50000 (38%)]    Loss: 0.096691  Acc: 96.000000
Train Epoch: 18 [25600/50000 (51%)]    Loss: 0.102257  Acc: 96.000000
Train Epoch: 18 [32000/50000 (64%)]    Loss: 0.103708  Acc: 96.000000
Train Epoch: 18 [38400/50000 (77%)]    Loss: 0.103074  Acc: 96.000000
Train Epoch: 18 [44800/50000 (90%)]    Loss: 0.106078  Acc: 96.000000
one epoch spend:  0:00:06.909887
EPOCH:18, ACC:74.86

Train Epoch: 19 [6400/50000 (13%)]    Loss: 0.074644  Acc: 97.000000
Train Epoch: 19 [12800/50000 (26%)]    Loss: 0.072871  Acc: 97.000000
Train Epoch: 19 [19200/50000 (38%)]    Loss: 0.075573  Acc: 97.000000
Train Epoch: 19 [25600/50000 (51%)]    Loss: 0.079646  Acc: 97.000000
Train Epoch: 19 [32000/50000 (64%)]    Loss: 0.081056  Acc: 97.000000
Train Epoch: 19 [38400/50000 (77%)]    Loss: 0.084256  Acc: 97.000000
Train Epoch: 19 [44800/50000 (90%)]    Loss: 0.086415  Acc: 97.000000
one epoch spend:  0:00:07.215059
EPOCH:19, ACC:75.69

Train Epoch: 20 [6400/50000 (13%)]    Loss: 0.062469  Acc: 97.000000
Train Epoch: 20 [12800/50000 (26%)]    Loss: 0.061595  Acc: 97.000000
Train Epoch: 20 [19200/50000 (38%)]    Loss: 0.062788  Acc: 97.000000
Train Epoch: 20 [25600/50000 (51%)]    Loss: 0.065734  Acc: 97.000000
Train Epoch: 20 [32000/50000 (64%)]    Loss: 0.067006  Acc: 97.000000
Train Epoch: 20 [38400/50000 (77%)]    Loss: 0.066818  Acc: 97.000000
Train Epoch: 20 [44800/50000 (90%)]    Loss: 0.068419  Acc: 97.000000
one epoch spend:  0:00:07.187726
EPOCH:20, ACC:74.23

CIFAR10 pytorch LeNet Train: EPOCH:20, BATCH_SZ:64, LR:0.01, ACC:75.89
train spend time:  0:02:30.334005

Process finished with exit code 0

准确率达到75%，对比LeNet-5的63%，有大幅提升。

上一篇：Javascript玩转继承（二）


下一篇：Masonry -- 使用纯代码进行iOS应用的autolayout自适应布局