GBlock Class Reference

Inheritance diagram for GBlock:

Collaboration diagram for GBlock:

Public Member Functions
def	__init__ (self, in_channels, out_channels, which_conv=SNConv2d, which_bn=bn, activation=None, upsample=None, channel_ratio=4)
	Constructor.
def	forward (self, x, y)
	forward

Public Attributes
	out_channels
	input channels
	hidden_channels
	hidden channels
	which_bn
	which convolution
	activation
	activation
	conv1
	conv1
	conv2
	conv2
	conv3
	conv3
	conv4
	conv4
	bn1
	bn1
	bn2
	bn2
	bn3
	bn3
	bn4
	bn4
	upsample
	upsample layers

Detailed Description

GBlock

Definition at line 990 of file ieagan.py.

Constructor & Destructor Documentation

__init__()

def __init__	(		self,
			in_channels,
			out_channels,
			which_conv = SNConv2d,
			which_bn = bn,
			activation = None,
			upsample = None,
			channel_ratio = 4
	)

Constructor.

Definition at line 994 of file ieagan.py.

    ):
        super(GBlock, self).__init__()
 
        
        self.in_channels, self.out_channels = in_channels, out_channels
        
        self.hidden_channels = self.in_channels // channel_ratio
        
        self.which_conv, self.which_bn = which_conv, which_bn
        
        self.activation = activation
        # Conv layers
        
        self.conv1 = self.which_conv(
            self.in_channels, self.hidden_channels, kernel_size=1, padding=0
        )
        
        self.conv2 = self.which_conv(self.hidden_channels, self.hidden_channels)
        
        self.conv3 = self.which_conv(self.hidden_channels, self.hidden_channels)
        
        self.conv4 = self.which_conv(
            self.hidden_channels, self.out_channels, kernel_size=1, padding=0
        )
        # Batchnorm layers
        
        self.bn1 = self.which_bn(self.in_channels)
        
        self.bn2 = self.which_bn(self.hidden_channels)
        
        self.bn3 = self.which_bn(self.hidden_channels)
        
        self.bn4 = self.which_bn(self.hidden_channels)
        
        self.upsample = upsample
 

Member Function Documentation

forward()

def forward	(		self,
			x,
			y
	)

forward

Definition at line 1040 of file ieagan.py.

    def forward(self, x, y):
        # Project down to channel ratio
        h = self.conv1(self.activation(self.bn1(x, y)))
        # Apply next BN-ReLU
        h = self.activation(self.bn2(h, y))
        # Drop channels in x if necessary
        if self.in_channels != self.out_channels:
            x = x[:, : self.out_channels]
        # Upsample both h and x at this point
        if self.upsample:
            h = self.upsample(h)
            x = self.upsample(x)
        # 3x3 convs
        h = self.conv2(h)
        h = self.conv3(self.activation(self.bn3(h, y)))
        # Final 1x1 conv
        h = self.conv4(self.activation(self.bn4(h, y)))
        return h + x
 
 

Member Data Documentation

activation

activation

activation

Definition at line 1013 of file ieagan.py.

bn1

bn1

bn1

Definition at line 1029 of file ieagan.py.

bn2

bn2

bn2

Definition at line 1031 of file ieagan.py.

bn3

bn3

bn3

Definition at line 1033 of file ieagan.py.

bn4

bn4

bn4

Definition at line 1035 of file ieagan.py.

conv1

conv1

conv1

Definition at line 1016 of file ieagan.py.

conv2

conv2

conv2

Definition at line 1020 of file ieagan.py.

conv3

conv3

conv3

Definition at line 1022 of file ieagan.py.

conv4

conv4

conv4

Definition at line 1024 of file ieagan.py.

hidden_channels

hidden_channels

hidden channels

Definition at line 1009 of file ieagan.py.

out_channels

out_channels

input channels

Definition at line 1007 of file ieagan.py.

upsample

upsample

upsample layers

Definition at line 1037 of file ieagan.py.

which_bn

which_bn

which convolution

Definition at line 1011 of file ieagan.py.

---

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

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