AFF：特征通道注意力融合.
特征融合(Feature Fusion)是指将来自不同层或不同分支的特征进行组合，一些常见的融合方法是求和(sum)或串联(concatenate)。本文作者提出了一种通过注意力机制进行特征融合的方法AFF，可以应用到skip connection、multi-path、FPN等需要进行特征融合的地方。
1. Multi-Scale Channel Attention Module (MS-CAM)
作者首先提出了一种多尺度通道注意力模块MS-CAM，该模块通过并行的通道注意力和逐点注意力重构输入特征：
Pytorch代码如下：
class MS_CAM(nn.Module):
def __init__(self, channels=64, r=4):
super(MS_CAM, self).__init__()
inter_channels = int(channels // r)
self.local_att = nn.Sequential(
nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(inter_channels),
nn.ReLU(inplace=True),
nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(channels),
)
self.global_att = nn.Sequential(
nn.AdaptiveAvgPool2d(1),
nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(inter_channels),
nn.ReLU(inplace=True),
nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(channels),
)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
xl = self.local_att(x)
xg = self.global_att(x)
xlg = xl + xg
wei = self.sigmoid(xlg)
return x * wei
2. Attentional Feature Fusion (AFF)
作者提出注意力融合模块AFF，将两个不同特征进行融合：
Pytorch代码如下：
class AFF(nn.Module):
def __init__(self, channels=64, r=4):
super(AFF, self).__init__()
inter_channels = int(channels // r)
self.local_att = nn.Sequential(
nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(inter_channels),
nn.ReLU(inplace=True),
nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(channels),
)
self.global_att = nn.Sequential(
nn.AdaptiveAvgPool2d(1),
nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(inter_channels),
nn.ReLU(inplace=True),
nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(channels),
)
self.sigmoid = nn.Sigmoid()
def forward(self, x, y):
xy = x + y
xl = self.local_att(xy)
xg = self.global_att(xy)
xlg = xl + xg
wei = self.sigmoid(xlg)
xo = x * wei + y * (1 - wei)
return xo
3. iterative Attentional Feature Fusion (iAFF)
作者提出迭代注意力融合模块iAFF，将两个不同特征进一步融合：
Pytorch代码如下：
class iAFF(nn.Module):
def __init__(self, channels=64, r=4):
super(iAFF, self).__init__()
inter_channels = int(channels // r)
# 本地注意力
self.local_att = nn.Sequential(
nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(inter_channels),
nn.ReLU(inplace=True),
nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(channels),
)
# 全局注意力
self.global_att = nn.Sequential(
nn.AdaptiveAvgPool2d(1),
nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(inter_channels),
nn.ReLU(inplace=True),
nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(channels),
)
# 第二次本地注意力
self.local_att2 = nn.Sequential(
nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(inter_channels),
nn.ReLU(inplace=True),
nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(channels),
)
# 第二次全局注意力
self.global_att2 = nn.Sequential(
nn.AdaptiveAvgPool2d(1),
nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(inter_channels),
nn.ReLU(inplace=True),
nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
nn.BatchNorm2d(channels),
)
self.sigmoid = nn.Sigmoid()
def forward(self, x, y):
xa = x + y
xl = self.local_att(xa)
xg = self.global_att(xa)
xlg = xl + xg
wei = self.sigmoid(xlg)
xi = x * wei + y * (1 - wei)
xl2 = self.local_att2(xi)
xg2 = self.global_att(xi)
xlg2 = xl2 + xg2
wei2 = self.sigmoid(xlg2)
xo = x * wei2 + y * (1 - wei2)
return xo
