CBAM Transformer 入门指南：从模型结构到实战应用

1次阅读

共计 3232 个字符，预计需要花费 9 分钟才能阅读完成。

Transformer 模型在 NLP 领域取得了巨大成功，但当它被应用到计算机视觉任务时，暴露出了一些问题。最主要的是传统 Transformer 缺乏对图像局部特征的关注能力，以及计算复杂度较高的问题。这导致模型在图像识别任务中表现不如预期，尤其在处理细粒度分类时效果欠佳。

CBAM(Convolutional Block Attention Module)通过引入通道注意力和空间注意力两个机制，有效改善了传统 Transformer 的不足。

通道注意力：让模型学习不同特征通道的重要性权重
空间注意力：让模型关注图像中的关键区域

这种双重注意力机制使得模型能够更有效地利用有限的参数和计算资源。

模型类型	计算复杂度	参数量	ImageNet 准确率
Vanilla Transformer	O(N^2)	大	78.2%
Swin Transformer	O(N)	中	81.3%
CBAM Transformer	O(N)	小	82.1%

下面是用 PyTorch 实现 CBAM 的关键代码：

import torch
import torch.nn as nn

class ChannelAttention(nn.Module):
    def __init__(self, in_planes, ratio=16):
        super(ChannelAttention, self).__init__()
        # 平均池化层
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        # 最大池化层
        self.max_pool = nn.AdaptiveMaxPool2d(1)

        # 全连接层 1
        self.fc1 = nn.Conv2d(in_planes, in_planes // ratio, 1, bias=False)
        self.relu1 = nn.ReLU()
        # 全连接层 2
        self.fc2 = nn.Conv2d(in_planes // ratio, in_planes, 1, bias=False)

        self.sigmoid = nn.Sigmoid()

    def forward(self, x):
        avg_out = self.fc2(self.relu1(self.fc1(self.avg_pool(x))))
        max_out = self.fc2(self.relu1(self.fc1(self.max_pool(x))))
        out = avg_out + max_out
        return self.sigmoid(out)

class SpatialAttention(nn.Module):
    def __init__(self, kernel_size=7):
        super(SpatialAttention, self).__init__()

        assert kernel_size in (3,7), 'kernel size must be 3 or 7'
        padding = 3 if kernel_size == 7 else 1

        self.conv1 = nn.Conv2d(2, 1, kernel_size, padding=padding, bias=False)
        self.sigmoid = nn.Sigmoid()

    def forward(self, x):
        avg_out = torch.mean(x, dim=1, keepdim=True)
        max_out, _ = torch.max(x, dim=1, keepdim=True)
        x = torch.cat([avg_out, max_out], dim=1)
        x = self.conv1(x)
        return self.sigmoid(x)

class CBAM(nn.Module):
    def __init__(self, in_planes, ratio=16, kernel_size=7):
        super(CBAM, self).__init__()
        self.ca = ChannelAttention(in_planes, ratio)
        self.sa = SpatialAttention(kernel_size)

    def forward(self, x):
        x = x * self.ca(x)
        x = x * self.sa(x)
        return x

下面展示如何将 CBAM 集成到 ResNet 中：

class ResNetWithCBAM(nn.Module):
    def __init__(self, block, layers, num_classes=1000):
        super(ResNetWithCBAM, self).__init__()
        self.inplanes = 64
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)

        # 在每个残差块后添加 CBAM 模块
        self.layer1 = self._make_layer(block, 64, layers[0])
        self.cbam1 = CBAM(256)

        self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
        self.cbam2 = CBAM(512)

        self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
        self.cbam3 = CBAM(1024)

        self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
        self.cbam4 = CBAM(2048)

        self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
        self.fc = nn.Linear(512 * block.expansion, num_classes)

    def _make_layer(self, block, planes, blocks, stride=1):
        # 标准 ResNet 的实现
        pass

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)

        x = self.layer1(x)
        x = self.cbam1(x)

        x = self.layer2(x)
        x = self.cbam2(x)

        x = self.layer3(x)
        x = self.cbam3(x)

        x = self.layer4(x)
        x = self.cbam4(x)

        x = self.avgpool(x)
        x = torch.flatten(x, 1)
        x = self.fc(x)

        return x