Pytorch:降低U-net模型中的参数

Question

有人能给我一些关于如何在下面的U-net实现中降低参数数量的提示吗。我在训练数据上遇到了过度拟合的问题，我想降低参数，看看它是否能提高验证数据的准确性。图层：

第一个2D

layers = [
nn.Conv2d(in_channels, middle_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(middle_channels),
nn.ReLU(inplace=True),
nn.Conv2d(middle_channels, out_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True)
]

编码器2D

layers = [
nn.MaxPool2d(kernel_size=downsample_kernel),
nn.Conv2d(in_channels, middle_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(middle_channels),
nn.ReLU(inplace=True),
nn.Conv2d(middle_channels, out_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True)
]

Center2D

layers = [
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(in_channels, middle_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(middle_channels),
nn.ReLU(inplace=True),
nn.Conv2d(middle_channels, out_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True),
nn.ConvTranspose2d(out_channels, deconv_channels, kernel_size=2, stride=2)
]

解码器2D

layers = [
nn.Conv2d(in_channels, middle_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(middle_channels),
nn.ReLU(inplace=True),
nn.Conv2d(middle_channels, out_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True),
nn.ConvTranspose2d(out_channels, deconv_channels, kernel_size=2, stride=2)
]

最后2D

layers = [
nn.Conv2d(in_channels, middle_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(middle_channels),
nn.ReLU(inplace=True),
nn.Conv2d(middle_channels, middle_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(middle_channels),
nn.ReLU(inplace=True),
nn.Conv2d(middle_channels, out_channels, kernel_size=1),
nn.Softmax(dim=1)
]

Answer 1

减少参数数量的一种方法是减少卷积中的通道数量。您将无法更改模型输入和输出通道的数量，因为它们取决于数据，但您可以更改中间通道的数量。

记住，一层的输出就是下一层的输入，所以对于每一对层，保持第一层中输出通道的数量与第二层中输入通道的数量相同。就是一个例子

layers = [
nn.Conv2d(in_channels, middle_channels//2, kernel_size=3, padding=1),
nn.BatchNorm2d(middle_channels//2),
nn.ReLU(inplace=True),
nn.Conv2d(middle_channels//2, out_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True)
]

现在，谈到过拟合的最初问题，首先，在减小模型大小之前，您可能想先尝试使用其他东西。有些事情包括数据扩充和丢弃。