网络模型各个层级输出如何计算

在PyTorch中，可以通过在网络模型中添加hook函数来获取每个层级的输出。hook函数是一个可以在网络中的指定层级的前向传播或后向传播过程中被调用的函数。

下面是一个示例，展示如何计算网络模型各个层级的输出：

import torch
import torch.nn as nn

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu1 = nn.ReLU(inplace=True)
        self.pool1 = nn.MaxPool2d(kernel_size=2, stride=2)
        self.conv2 = nn.Conv2d(64, 128, kernel_size=3, stride=1, padding=1)
        self.bn2 = nn.BatchNorm2d(128)
        self.relu2 = nn.ReLU(inplace=True)
        self.pool2 = nn.MaxPool2d(kernel_size=2, stride=2)
        self.conv3 = nn.Conv2d(128, 256, kernel_size=3, stride=1, padding=1)
        self.bn3 = nn.BatchNorm2d(256)
        self.relu3 = nn.ReLU(inplace=True)
        self.pool3 = nn.MaxPool2d(kernel_size=2, stride=2)

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu1(x)
        x = self.pool1(x)
        x = self.conv2(x)
        x = self.bn2(x)
        x = self.relu2(x)
        x = self.pool2(x)
        x = self.conv3(x)
        x = self.bn3(x)
        x = self.relu3(x)
        x = self.pool3(x)
        return x

def print_output_shape(module, input, output):
    print("{} output shape: {}".format(module.__class__.__name__, output.size()))

net = Net()
hook_handles = []
for layer in net.children():
    handle = layer.register_forward_hook(print_output_shape)
    hook_handles.append(handle)

x = torch.randn(1, 3, 32, 32)
_ = net(x)

for handle in hook_handles:
    handle.remove()

在上面的代码中，我们定义了一个包含三个卷积层和池化层的卷积神经网络。然后，我们通过在每个卷积层和池化层中添加hook函数，打印出了它们的输出形状。最后，我们从每个层级中删除了hook函数。

当我们运行这段代码时，它会输出每个层级的输出形状，类似于以下内容：

Conv2d output shape: torch.Size([1, 64, 32, 32])
BatchNorm2d output shape: torch.Size([1, 64, 32, 32])
ReLU output shape: torch.Size([1, 64, 32, 32])
MaxPool2d output shape: torch.Size([1, 64, 16, 16])
Conv2d output shape: torch.Size([1, 128, 16, 16])
BatchNorm2d output shape: torch.Size([1, 128, 16, 16])
ReLU output shape: torch.Size([1, 128, 16, 16])
MaxPool2d output shape: torch.Size([1, 128, 8, 8])
Conv2d output shape: torch.Size([1, 256, 8, 8])
BatchNorm2d output shape: torch.Size([1, 256, 8, 8])
ReLU output shape: torch.Size([1, 256, 8, 8])
MaxPool2d output shape: torch.Size([1, 256, 4, 4])

可以看到，我们成功地获取了每个层级的输出形状。