CNN 2 LeNet5, AlexNet, MyCNN,ResNet 比较

承接上一篇 CNN 1 对其他CNN结构进行比较。

LeNet5   time   82.2345199584961    AlexNet   time   215.45081973075867    MyCNN   time   486.6250228881836    ResNet   time   662.9733195304871    
LeNet5   loss   1.575899446105957    AlexNet   loss   1.3081466011047362    MyCNN   loss   1.0375914037704468    ResNet   loss   1.2145104179382324    
LeNet5   acc   0.424699991941452    AlexNet   acc   0.5227000117301941    MyCNN   acc   0.6352999806404114    ResNet   acc   0.5829      

自己参照设计的CNN网络精确度最高，对比一下各个网络参数分析：

MyCNN 参数
Total params: 2,316,170
Trainable params: 2,314,826
Non-trainable params: 1,344

lenet5
Total params: 15,699
Trainable params: 15,699
Non-trainable params: 0

alexnet
Total params: 17,369,022
Trainable params: 17,369,022
Non-trainable params: 0

resnet
Total params: 3,638
Trainable params: 3,296
Non-trainable params: 342

待：
使用 存储好的模型去预测图片。