CBAM_attention Class Reference

Inheritance diagram for CBAM_attention:

Public Member Functions
	__init__ (self, channels, which_conv=SNConv2d, reduction=8, attention_kernel_size=3)
	Constructor.
	forward (self, x)
	forward

Public Attributes
	avg_pool = nn.AdaptiveAvgPool2d(1)
	average pooling
	max_pool = nn.AdaptiveMaxPool2d(1)
	max pooling
	fc1
	fcl
	relu = nn.ReLU(inplace=True)
	relu
	fc2
	f2c
	sigmoid_channel = nn.Sigmoid()
	sigmoid channel
	conv_after_concat
	convolution after concatenation
	sigmoid_spatial = nn.Sigmoid()
	sigmoid_spatial

Detailed Description

CBAM attention

Definition at line 683 of file ieagan.py.

Constructor & Destructor Documentation

__init__()

__init__	(		self,
			channels,
			which_conv = SNConv2d,
			reduction = 8,
			attention_kernel_size = 3 )

Constructor.

Definition at line 687 of file ieagan.py.

    ):
        super(CBAM_attention, self).__init__()
        
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        
        self.max_pool = nn.AdaptiveMaxPool2d(1)
        
        self.fc1 = which_conv(
            channels, channels // reduction, kernel_size=1, padding=0
        )
        
        self.relu = nn.ReLU(inplace=True)
        
        self.fc2 = which_conv(
            channels // reduction, channels, kernel_size=1, padding=0
        )
        
        self.sigmoid_channel = nn.Sigmoid()
        
        self.conv_after_concat = which_conv(
            2,
            1,
            kernel_size=attention_kernel_size,
            stride=1,
            padding=attention_kernel_size // 2,
        )
        
        self.sigmoid_spatial = nn.Sigmoid()
 

Member Function Documentation

forward()

forward	(		self,
			x )

forward

Definition at line 723 of file ieagan.py.

    def forward(self, x):
        # Channel attention module
        module_input = x
        avg = self.avg_pool(x)
        mx = self.max_pool(x)
        avg = self.fc1(avg)
        mx = self.fc1(mx)
        avg = self.relu(avg)
        mx = self.relu(mx)
        avg = self.fc2(avg)
        mx = self.fc2(mx)
        x = avg + mx
        x = self.sigmoid_channel(x)
        # Spatial attention module
        x = module_input * x
        module_input = x
        # b, c, h, w = x.size()
        avg = torch.mean(x, 1, True)
        mx, _ = torch.max(x, 1, True)
        x = torch.cat((avg, mx), 1)
        x = self.conv_after_concat(x)
        x = self.sigmoid_spatial(x)
        x = module_input * x
        return x
 
 

Member Data Documentation

avg_pool

avg_pool = nn.AdaptiveAvgPool2d(1)

average pooling

Definition at line 696 of file ieagan.py.

conv_after_concat

conv_after_concat

Initial value:

=  which_conv(
            2,
            1,
            kernel_size=attention_kernel_size,
            stride=1,
            padding=attention_kernel_size // 2,
        )

convolution after concatenation

Definition at line 712 of file ieagan.py.

fc1

fc1

Initial value:

=  which_conv(
            channels, channels // reduction, kernel_size=1, padding=0
        )

fcl

Definition at line 700 of file ieagan.py.

fc2

fc2

Initial value:

=  which_conv(
            channels // reduction, channels, kernel_size=1, padding=0
        )

f2c

Definition at line 706 of file ieagan.py.

max_pool

max_pool = nn.AdaptiveMaxPool2d(1)

max pooling

Definition at line 698 of file ieagan.py.

relu

relu = nn.ReLU(inplace=True)

relu

Definition at line 704 of file ieagan.py.

sigmoid_channel

sigmoid_channel = nn.Sigmoid()

sigmoid channel

Definition at line 710 of file ieagan.py.

sigmoid_spatial

sigmoid_spatial = nn.Sigmoid()

sigmoid_spatial

Definition at line 720 of file ieagan.py.

---

The documentation for this class was generated from the following file:

• pxd/scripts/pxd/background_generator/models/ieagan.py