ResidualBlock Class Reference

Class for the residual block layer. More...

Inheritance diagram for ResidualBlock:

Collaboration diagram for ResidualBlock:

Public Member Functions
def	__init__ (self, ninput, noutput, upsample=True)
	Constructor to create a new residual block layer.
def	forward (self, x)
	Function to perform a forward pass.

Public Attributes
	upsample
	Whether to double the height and width of input.
	conv
	Convolutional layer in the shortcut branch.
	norm1
	First batch normalization layer in the residual branch.
	conv1
	First convolutional layer in the residual branch.
	norm2
	Second batch normalization layer in the residual branch.
	conv2
	Second convolutional layer in the residual branch.

Detailed Description

Class for the residual block layer.

Residual block layer.

Definition at line 19 of file resnet.py.

Constructor & Destructor Documentation

__init__()

def __init__	(		self,
			ninput,
			noutput,
			upsample = True
	)

Constructor to create a new residual block layer.

Definition at line 24 of file resnet.py.

    def __init__(self, ninput, noutput, upsample=True):
        super().__init__()
        self.upsample = upsample
        # shortcut branch
        self.conv = None
        if upsample or (ninput != noutput):
            self.conv = nn.Conv2d(ninput, noutput, 1, 1, 0)
        # residual branch
        self.norm1 = nn.BatchNorm2d(ninput)
        self.conv1 = nn.Conv2d(ninput, noutput, 3, 1, 1)
        self.norm2 = nn.BatchNorm2d(noutput)
        self.conv2 = nn.Conv2d(noutput, noutput, 3, 1, 1)
 

Member Function Documentation

forward()

def forward	(		self,
			x
	)

Function to perform a forward pass.

Compute the layer output for a given input.

Definition at line 39 of file resnet.py.

    def forward(self, x):
        """Compute the layer output for a given input."""
        # residual branch
        h = x
        h = self.norm1(h)
        h.relu_()
        if self.upsample:
            h = F.interpolate(h, mode="nearest", scale_factor=2)
        h = self.conv1(h)
        h = self.norm2(h)
        h.relu_()
        h = self.conv2(h)
        # shortcut branch
        if self.upsample:
            x = F.interpolate(x, mode="nearest", scale_factor=2)
        if self.conv:
            x = self.conv(x)
        # return sum of both
        return h + x
 

Member Data Documentation

conv

conv

Convolutional layer in the shortcut branch.

Definition at line 28 of file resnet.py.

conv1

conv1

First convolutional layer in the residual branch.

Definition at line 33 of file resnet.py.

conv2

conv2

Second convolutional layer in the residual branch.

Definition at line 35 of file resnet.py.

norm1

norm1

First batch normalization layer in the residual branch.

Definition at line 32 of file resnet.py.

norm2

norm2

Second batch normalization layer in the residual branch.

Definition at line 34 of file resnet.py.

upsample

upsample

Whether to double the height and width of input.

Definition at line 26 of file resnet.py.

---

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

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