pytorch中children()modules()，named_children()，named_modules()，named_parameters()，parameters()的使用

àì夳堔傛蜴生んèń 2023-02-10 03:44 74阅读 0赞

class SimpleCNN(nn.Module):  
    def \_\_init\_\_(self):  
        super(SimpleCNN, self).\_\_init\_\_()  
        layer1 = nn.Sequential()  
        \#class torch.nn.Conv2d(in\_channels, out\_channels, kernel\_size, stride=1, padding=0, dilation=1, groups=1, bias=True)  
        layer1.add\_module('conv1', nn.Conv2d(3, 32, 3, 1, padding=1))\#b, 32, 32,32   \#输入数据体的深度;out channels 表示输出数据体的深度;  
        layer1.add\_module('relu1', nn.ReLU(True))  
        layer1.add\_module('pool1', nn.MaxPool2d(2,2))\#b, 32, 16,16  
        self.layer1 = layer1  
          
        layer2 = nn.Sequential()  
        layer2.add\_module('conv2', nn.Conv2d(32, 64, 3, 1, padding=1))\#b, 64, 16,16  
        layer2.add\_module('relu2', nn.ReLU(True))  
        layer2.add\_module('pool2', nn.MaxPool2d(2,2))\#b, 64, 8,8  
        self.layer2 = layer2  
          
        layer3 = nn.Sequential()  
        layer3.add\_module('conv3', nn.Conv2d(64, 128, 3, 1, padding=1))\#b, 128, 8,8  
        layer3.add\_module('relu3', nn.ReLU(True))  
        layer3.add\_module('pool3', nn.MaxPool2d(2,2))\#b, 128, 4,4  
        self.layer3 = layer3  
          
        layer4 = nn.Sequential()  
        layer4.add\_module('fc1', nn.Linear(2048,512))  
        layer4.add\_module('fc1\_relu1', nn.ReLU(True))  
        layer4.add\_module('fc2', nn.Linear(512,64))  
        layer4.add\_module('fc2\_relu2', nn.ReLU(True))  
        layer4.add\_module('fc3', nn.Linear(64,10))  
        self.layer4 = layer4  
          
        def forward(self, x):  
            conv1 = self.layer1(x)  
            conv2 = self.layer2(conv1)  
            conv3 = self.layer3(conv2)  
            print('conv3:\{\}'.format(conv3.size()))  
            """  
            卷积或者池化之后的tensor的维度为(batchsize，channels，x，y)，  
            其中x.size(0)指batchsize的值，最后通过x.view(x.size(0), -1)将tensor的结构转换为了(batchsize, channels\*x\*y)，  
            即将（channels，x，y）拉直，然后就可以和fc层连接了

"""  
            fc\_input = conv3.view(conv3.size(0), -1)  
            out = self.layer4(fc\_input)  
            return out

model = SimpleCNN()  
print(model)

new\_model = nn.Sequential(\*list(model.children()))\#这个会返回下一级模块的选代器  
print(new\_model)

new\_model = nn.Sequential(\*list(model.modules()))\#返回模型中所有模块的迭代器，  
print(new\_model)

\#这两个不仅返回模块的迭代器，还返问网络层的名子  
for name, module in model.named\_children():  
    print('children name:', name)  
    print('children module:', module)  
print('===========================')  
for name, module in model.named\_modules():  
    print('children name:', name)  
    print('children module:', module)

\#named\_parameters()是给出网络的名字和参数的迭代器， parameters()会给出一个网络的全部参数的选代器

for para in model.named\_parameters():  
      
    print(para\[0\])

for para in model.parameters():  
      
    print(para\[0\])