Attention Class Reference

Inheritance diagram for Attention:

Public Member Functions
def	__init__ (self, ch, which_conv=SNConv2d)
	Constructor.
def	forward (self, x)
	forward

Public Attributes
	ch
	Channel multiplier.
	which_conv
	which_conv
	theta
	theta
	phi
	phi
	g
	g
	o
	o
	gamma
	Learnable gain parameter.

Detailed Description

Attention

Definition at line 749 of file ieagan.py.

Constructor & Destructor Documentation

__init__()

def __init__	(		self,
			ch,
			which_conv = SNConv2d
	)

Constructor.

Definition at line 753 of file ieagan.py.

    def __init__(self, ch, which_conv=SNConv2d):
        super(Attention, self).__init__()
        
        self.ch = ch
        
        self.which_conv = which_conv
        
        self.theta = self.which_conv(
            self.ch, self.ch // 8, kernel_size=1, padding=0, bias=False
        )
        
        self.phi = self.which_conv(
            self.ch, self.ch // 8, kernel_size=1, padding=0, bias=False
        )
        
        self.g = self.which_conv(
            self.ch, self.ch // 2, kernel_size=1, padding=0, bias=False
        )
        
        self.o = self.which_conv(
            self.ch // 2, self.ch, kernel_size=1, padding=0, bias=False
        )
        
        self.gamma = P(torch.tensor(0.0), requires_grad=True)
 

Member Function Documentation

forward()

def forward	(		self,
			x
	)

forward

Definition at line 779 of file ieagan.py.

    def forward(self, x):
        # Apply convs
        theta = self.theta(x)
        phi = F.max_pool2d(self.phi(x), [2, 2])
        g = F.max_pool2d(self.g(x), [2, 2])
        # Perform reshapes
        theta = theta.view(-1, self.ch // 8, x.shape[2] * x.shape[3])
        phi = phi.view(-1, self.ch // 8, x.shape[2] * x.shape[3] // 4)
        g = g.view(-1, self.ch // 2, x.shape[2] * x.shape[3] // 4)
        # Matmul and softmax to get attention maps
        beta = F.softmax(torch.bmm(theta.transpose(1, 2), phi), -1)
        # Attention map times g path
        o = self.o(
            torch.bmm(g, beta.transpose(1, 2)).view(
                -1, self.ch // 2, x.shape[2], x.shape[3]
            )
        )
        return self.gamma * o + x
 
 

Member Data Documentation

ch

ch

Channel multiplier.

Definition at line 756 of file ieagan.py.

g

g

g

Definition at line 768 of file ieagan.py.

gamma

gamma

Learnable gain parameter.

Definition at line 776 of file ieagan.py.

o

o

o

Definition at line 772 of file ieagan.py.

phi

phi

phi

Definition at line 764 of file ieagan.py.

theta

theta

theta

Definition at line 760 of file ieagan.py.

which_conv

which_conv

which_conv

Definition at line 758 of file ieagan.py.

---

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

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