pytorch 模型重写或嵌入

1.重写，假设模型名叫alexnet，除了init、forward方法外，还包含若干个函数

创建一个model类，去继承alexnet

代码示例

class model(alexnet):
    def __init__(self):
        super(model,self).__init__()
        #alexnet里的self变量继承

    def forward(self,x):
        x = self.func(x)
        return x

2.嵌入，将alexnet模型嵌入到model模型中

from torch.nn import *

class model(nn.Module):
    def __init__(self):
        super(model,self).__init__()
        self.head = nn.Conv2d(1,1,1)
        self.alexnet = models.alexnet(pretrained=True) # func 1 
        for k,v in alexnet._modules.items():  # func2
            key = 'self.' + k
            globals()[key] = v
        self.detect = nn.Linear(1000,10)

    def forward(self,x):
        x = self.head(x)
        x = self.alexnet(x)
        x = x.flatten()
        x = self.detect(x)
        return x

globals()[k]  可将字符串k转换为变量k