如何在其他两个卷积层之间添加一个卷积层？

原理是新的中间层的输入通道数应该等于上一层的输出通道数，新层的输出通道数应该等于下一层的输入通道数层：

class LeNet5(torch.nn.Module):
    def __init__(self):
        super(LeNet5, self).__init__()

        self.conv1 = torch.nn.Conv2d(
            in_channels=1, out_channels=6, kernel_size=5, padding=2)
        #######
        self.conv1_1 = torch.nn.Conv2d(
            in_channels=6, out_channels=6, kernel_size=5, padding=2)
        #######
        self.act1  = torch.nn.Tanh()
        self.pool1 = torch.nn.AvgPool2d(kernel_size=2, stride=2)

        self.pool1 = torch.nn.AvgPool2d(kernel_size=2, stride=2)


        self.conv2 = torch.nn.Conv2d(
            in_channels=6, out_channels=16, kernel_size=5, padding=0)
        self.act2  = torch.nn.Tanh()
        self.pool2 = torch.nn.AvgPool2d(kernel_size=2, stride=2)

        self.fc1   = torch.nn.Linear(5 * 5 * 16, 120)
        self.act3  = torch.nn.Tanh()

        self.fc2   = torch.nn.Linear(120, 84)
        self.act4  = torch.nn.Tanh()

        self.fc3   = torch.nn.Linear(84, 10)

    def forward(self, x):

        x = self.conv1(x)
        x = self.act1(x)
        #######
        x = self.conv1_1(x)
        x = self.act1(x)
        #######
        x = self.pool1(x)

        x = self.conv2(x)
        x = self.act2(x)
        x = self.pool2(x)

        x = x.view(x.size(0), x.size(1) * x.size(2) * x.size(3))

        x = self.fc1(x)
        x = self.act3(x)
        x = self.fc2(x)
        x = self.act4(x)
        x = self.fc3(x)

        return x

或者，犯错误。运行脚本时输出和输入层数的不一致会给出异常，指示必须组合的数字。

例如，

RuntimeError: running_mean 应该包含 768 个元素而不是 32

为了

   nn.Conv2d(48, **768**, 2, 1, 1), nn.BatchNorm2d(**32**), nn.ReLU(True),  

或者

RuntimeError: 给定组=1，大小为 [6, 512, 2, 2] 的权重，预期输入 [300, 768, 33, 9] 有 512 个通道，但有 768 个通道

    nn.Conv2d(48, 768, 2, 1, 1), nn.BatchNorm2d(768), nn.ReLU(True), 

    nn.Conv2d(512, 6, 2, 1), nn.ReLU(True), nn.MaxPool2d(2, 2),

在 SO 上，他们经常引用一个考虑通道、填充和跨步的公式，

也使用模型可视化

ZZNet(
  (cnn): Sequential(
    (0): Conv2d(3, 3, kernel_size=(1, 1), stride=(1, 1), padding=(1, 1))
    (1): ReLU(inplace=True)
    (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (3): Conv2d(3, 48, kernel_size=(2, 2), stride=(1, 1))
    (4): ReLU(inplace=True)
    (5): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (6): Conv2d(48, 768, kernel_size=(2, 2), stride=(1, 1), padding=(1, 1))
    (7): BatchNorm2d(768, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (8): ReLU(inplace=True)
    (9): Conv2d(512, 6, kernel_size=(2, 2), stride=(1, 1))
    (10): ReLU(inplace=True)
    (11): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (12): Conv2d(6, 6, kernel_size=(2, 2), stride=(1, 1), padding=(1, 1))
    (13): BatchNorm2d(6, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (14): ReLU(inplace=True)
    (15): Conv2d(6, 6, kernel_size=(2, 2), stride=(1, 1))
    (16): ReLU(inplace=True)
    (17): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (18): Conv2d(6, 6, kernel_size=(2, 2), stride=(1, 1), padding=(1, 1))
    (19): BatchNorm2d(6, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (20): ReLU(inplace=True)
    (21): Conv2d(6, 6, kernel_size=(2, 2), stride=(1, 1))
    (22): ReLU(inplace=True)
    (23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (24): Conv2d(6, 6, kernel_size=(2, 2), stride=(1, 1), padding=(1, 1))
    (25): BatchNorm2d(6, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (26): ReLU(inplace=True)
    (27): LeakyReLU(negative_slope=0.1)
    (28): Flatten()
    (29): Linear(in_features=60, out_features=60, bias=True)
    (30): LeakyReLU(negative_slope=0.1)
    (31): Linear(in_features=60, out_features=12, bias=True)
  )
)

更新，渲染层

net = ZZNet().to(device)

#1
print(net)

#2
# Print model's state_dict
print("Model's state_dict:")
for param_tensor in net.state_dict():
    print(param_tensor, "\t", net.state_dict()[param_tensor].size())


#3
torchstat.stat(net, (3, 32, 128))

#4 - отброшенные варианты
# for idx, m in enumerate(net.modules()):
#         print(idx, '->', m, type(m))
#
# # for name, param in net.parameters():
# #     if name in ['bias']:
# #         print('param: ', param.size())
#
# print('NN: ',net.parameters())
# for param in net.parameters():
#     print(type(param.data), param.size())