pytorch查看模型每一层的名称

1 model._modules.items()

model._modules.items()是一个包含模型所有子模块的迭代器。在PyTorch中，当我们定义一个nn.Module子类时，我们可以使用nn.Sequential或nn.ModuleDict等容器类将多个子模块组合成一个整体。在这种情况下，我们可以通过访问nn.Module类中的_modules属性来访问这些子模块。_modules是一个有序字典，其中键是子模块的名称，值是子模块对象

import torch.nn as nn

class MyModel(nn.Module):
    def __init__(self):
        super(MyModel, self).__init__()
        self.conv1 = nn.Conv2d(3, 16, kernel_size=3, stride=1, padding=1)
        self.conv2 = nn.Conv2d(16, 32, kernel_size=3, stride=1, padding=1)
        self.relu = nn.ReLU()

    def forward(self, x):
        x = self.conv1(x)
        x = self.relu(x)
        x = self.conv2(x)
        x = self.relu(x)
        return x

model = MyModel()
for name, module in model._modules.items():
    print(name, module)

输出如下:

conv1 Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
conv2 Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
relu ReLU()

在上面的例子中，我们可以通过迭代model._modules.items()来访问模型中的每个子模块。name是子模块的名称（例如"conv1"、“conv2"和"relu”），module是子模块的实例对象（例如nn.Conv2d和nn.ReLU）。在递归模型中，每个子模块都可以是一个递归模型，因此我们可以使用_modules属性来递归遍历整个模型的所有子模块。

 2 named_modules()

import torch.nn as nn

class MyModel(nn.Module):
    def __init__(self):
        super(MyModel, self).__init__()
        self.conv1 = nn.Conv2d(3, 16, kernel_size=3, stride=1, padding=1)
        self.conv2 = nn.Conv2d(16, 32, kernel_size=3, stride=1, padding=1)
        self.relu = nn.ReLU()

    def forward(self, x):
        x = self.conv1(x)
        x = self.relu(x)
        x = self.conv2(x)
        x = self.relu(x)
        return x

model = MyModel()
for name, module in model.named_modules():
    print(name, module)

 

 MyModel(
  (conv1): Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv2): Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu): ReLU()
)
conv1 Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
conv2 Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
relu ReLU()

3 查看每一层的参数名和参数

import torch.nn as nn

class MyModel(nn.Module):
    def __init__(self):
        super(MyModel, self).__init__()
        self.conv1 = nn.Conv2d(3, 16, kernel_size=3, stride=1, padding=1)
        self.conv2 = nn.Conv2d(16, 32, kernel_size=3, stride=1, padding=1)
        self.relu = nn.ReLU()

    def forward(self, x):
        x = self.conv1(x)
        x = self.relu(x)
        x = self.conv2(x)
        x = self.relu(x)
        return x

model = MyModel()


for name, param in model.named_parameters():
    print(name, param.shape)

conv1.weight torch.Size([16, 3, 3, 3])
conv1.bias torch.Size([16])
conv2.weight torch.Size([32, 16, 3, 3])
conv2.bias torch.Size([32])

4 named_children() 和  named_modules()

import torch.nn as nn

class MyModel(nn.Module):
    def __init__(self):
        super(MyModel, self).__init__()
        self.conv1 = nn.Conv2d(3, 16, kernel_size=3, stride=1, padding=1)
        self.conv2 = nn.Conv2d(16, 32, kernel_size=3, stride=1, padding=1)
        self.relu = nn.ReLU()
        self.layer3 = nn.Sequential(
            nn.Linear(32,10),
            nn.ReLU(inplace=True)
        )

    def forward(self, x):
        x = self.conv1(x)
        x = self.relu(x)
        x = self.conv2(x)
        x = self.relu(x)
        x = self.layer3(x)
        return x

model = MyModel()


print("named_children:")
for name, module in model.named_children():
    print(name, module)

print("--------------------------------------------\n")
print("named_modules:")
for name, module in model.named_modules():
    print(name, module)

named_children:
conv1 Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
conv2 Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
relu ReLU()
layer3 Sequential(
  (0): Linear(in_features=32, out_features=10, bias=True)
  (1): ReLU(inplace=True)
)
--------------------------------------------

named_modules:
 MyModel(
  (conv1): Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv2): Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu): ReLU()
  (layer3): Sequential(
    (0): Linear(in_features=32, out_features=10, bias=True)
    (1): ReLU(inplace=True)
  )
)
conv1 Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
conv2 Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
relu ReLU()
layer3 Sequential(
  (0): Linear(in_features=32, out_features=10, bias=True)
  (1): ReLU(inplace=True)
)
layer3.0 Linear(in_features=32, out_features=10, bias=True)
layer3.1 ReLU(inplace=True)

5 modules()

函数返回模型的所有组成模块，包括子模块的组成模块.

import torch.nn as nn

class MyModel(nn.Module):
    def __init__(self):
        super(MyModel, self).__init__()
        self.conv1 = nn.Conv2d(3, 16, kernel_size=3, stride=1, padding=1)
        self.conv2 = nn.Conv2d(16, 32, kernel_size=3, stride=1, padding=1)
        self.relu = nn.ReLU()
        self.layer3 = nn.Sequential(
            nn.Linear(32,10),
            nn.ReLU(inplace=True)
        )

    def forward(self, x):
        x = self.conv1(x)
        x = self.relu(x)
        x = self.conv2(x)
        x = self.relu(x)
        x = self.layer3(x)
        return x

model = MyModel()



for module in model.modules():
    print(module)

MyModel(
  (conv1): Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv2): Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu): ReLU()
  (layer3): Sequential(
    (0): Linear(in_features=32, out_features=10, bias=True)
    (1): ReLU(inplace=True)
  )
)
Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU()
Sequential(
  (0): Linear(in_features=32, out_features=10, bias=True)
  (1): ReLU(inplace=True)
)
Linear(in_features=3

6 named_children() and named_parameters()

import torch.nn as nn

class MyModel(nn.Module):
    def __init__(self):
        super(MyModel, self).__init__()
        self.conv1 = nn.Conv2d(3, 16, kernel_size=3, stride=1, padding=1)
        self.conv2 = nn.Conv2d(16, 32, kernel_size=3, stride=1, padding=1)
        self.relu = nn.ReLU()
        self.layer3 = nn.Sequential(
            nn.Linear(32,10),
            nn.Conv2d(10, 10, kernel_size=3, stride=1, padding=1)

        )

    def forward(self, x):
        x = self.conv1(x)
        x = self.relu(x)
        x = self.conv2(x)
        x = self.relu(x)
        x = self.layer3(x)
        return x

model = MyModel()


for name, child_modle in model.named_children():
    print(name, ":")
    # print(name, child_modle, ":")
    for name, param in child_modle.named_parameters():
        print("    ", name, param.shape)

conv1 :
     weight torch.Size([16, 3, 3, 3])
     bias torch.Size([16])
conv2 :
     weight torch.Size([32, 16, 3, 3])
     bias torch.Size([32])
relu :
layer3 :
     0.weight torch.Size([10, 32])
     0.bias torch.Size([10])
     1.weight torch.Size([10, 10, 3, 3])
     1.bias torch.Size([10])