resnet18加入ema注意力机制

# 在ResNet18中嵌入EMA（Exponential Moving Average）注意力机制的示例代码如下：
import torch
import torch.nn as nn
import torch.nn.functional as F

class EMAAttention(nn.Module):
    def __init__(self, in_channels, reduction_ratio=16):
        super(EMAAttention, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.fc = nn.Sequential(
            nn.Linear(in_channels, in_channels // reduction_ratio),
            nn.ReLU(inplace=True),
            nn.Linear(in_channels // reduction_ratio, in_channels),
            nn.Sigmoid()
        )
        self.register_buffer('ema', torch.zeros(1))

    def forward(self, x):
        b, c, _, _ = x.size()
        y = self.avg_pool(x).view(b, c)
        y = self.fc(y).view(b, c, 1, 1)
        self.ema = 0.1 * y.mean((0, 2, 3)) + 0.9 * self.ema
        return F.relu(x * self.ema.view(1, c, 1, 1))

# 在ResNet18中使用EMA注意力机制
class BasicBlock(nn.Module):
    def __init__(self, in_planes, planes, stride=1, downsample=None):
        super(BasicBlock, self).__init__()
        self.ema_attention = EMAAttention(in_planes)
        # 其他部分省略

    def forward(self, x):
        # 其他部分省略
        out = self.ema_attention(out)
        # 其他部分省略