yolov8添加CBAM

要在YOLOv8中添加CBAM(Convolutional Block Attention Module),首先需要导入相关的库，然后在YOLOv8的网络结构中添加CBAM模块。以下是一个简单的示例： 1. 导入相关库： ```python import torch import torch.nn as nn import torch.nn.functional as F ``` 2. 定义CBAM模块： ```python class ChannelAttention(nn.Module): def __init__(self, in_planes, ratio=16): super(ChannelAttention, self).__init__() self.avg_pool = nn.AdaptiveAvgPool2d(1) self.max_pool = nn.AdaptiveMaxPool2d(1) self.fc1 = nn.Conv2d(in_planes, in_planes // ratio, 1, bias=False) self.relu1 = nn.ReLU() 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) ``` 3. 在YOLOv8的网络结构中添加CBAM模块： ```python class YOLOv8(nn.Module): def __init__(self): super(YOLOv8, self).__init__() # ... 其他层 ... self.cbam1 = ChannelAttention() self.cbam2 = ChannelAttention() # ... 其他层 ... ``` 4. 在前向传播过程中使用CBAM模块： ```python def forward(self, x): x = self.conv1(x) # C_{in} x H_{in/2} x W_{in/2} -> C_{out} x H_{out} x W_{out} x = self.relu1(self.bn1(x)) # C_{out} x H_{out} x W_{out} -> C_{out} x H_{out} x W_{out} x = self.cbam1(x) # C_{out} x H_{out} x W_{out} -> C_{out} x H_{out} x W_{out} with C_{out}/2 channels after attention and before residual connection x = self.residual(x) # C_{out} x H_{out} x W_{out} -> C_{out} x H_{out} x W_{out} with skip connection and residual connection x = self.relu2(self.bn2(x)) # C_{out} x H_{out} x W_{out} -> C_{out} x H_{out} x W_{out} with a shortcut connection to the skip connection path of the residual block above it (the identity mapping is applied here). This is used to reduce computational complexity when training deep networks with large numbers of parameters (e.g., ResNet-50). The shortcut connection is also known as a "skip connection" because the output of the layer isirectly passed (without

### 在YOLOv8中集成CBAM模块 为了在YOLOv8中集成CBAM模块以增强模型性能，可以借鉴先前版本的经验并作出适当调整。通常有两种主要方法来引入CBAM模块： #### 方法一：SPPF层前添加CBAM 可以在主干网络的SPPF（Spatial Pyramid Pooling-Fast）层之前加入一层CBAM模块[^1]。 ```yaml # yolov8s_CBAM.yaml backbone: ... [[-1, 'CBAM', {c1: $nc}], # 添加CBAM模块于SPPF之前 [-1, 'SPPF', {...}] ``` 此配置文件片段展示了如何修改YOLOv8的YAML配置，在指定位置插入CBAM模块。 #### 方法二：替换C3层为CBAM 另一种策略是将Backbone中的所有C3层替换成CBAM模块。 ```yaml # yolov8s_CBAM.yaml backbone: ... [[...], ['CBAM', {c1: ...}]] # 替换原有的C3层结构 ``` 对于具体的Python实现，基于已有的`class CBAM(nn.Module)`定义[^3]，可以直接利用该类作为构建基础，并将其嵌入到YOLOv8架构之中。 ```python from models.common import Conv, BottleneckCSP import torch.nn as nn class ChannelAttention(nn.Module): ... class SpatialAttention(nn.Module): ... class CBAM(nn.Module): """Convolutional Block Attention Module.""" def __init__(self, c1, kernel_size=7): super().__init__() self.channel_attention = ChannelAttention(c1) self.spatial_attention = SpatialAttention(kernel_size) def forward(self, x): return self.spatial_attention(self.channel_attention(x)) ``` 通过上述代码示例可以看出，CBAM由两个子组件构成——通道注意力机制和空间注意力机制；这两个部分共同作用以提升特征表达能力[^2]。