loki2.models.utils.blocks

Building blocks for neural network architectures.

This module provides reusable convolutional and deconvolutional blocks for use in segmentation networks.

Module Contents

class loki2.models.utils.blocks.Conv2DBlock(in_channels: int, out_channels: int, kernel_size: int = 3, dropout: float = 0)

Bases: torch.nn.Module

Conv2D block with convolution, batch normalization, ReLU, and dropout.

Parameters:

• in_channels – Number of input channels for convolution.

• out_channels – Number of output channels for convolution.

• kernel_size – Kernel size for convolution. Defaults to 3.

• dropout – Dropout rate. Defaults to 0.

block

forward(x)

class loki2.models.utils.blocks.Deconv2DBlock(in_channels: int, out_channels: int, kernel_size: int = 3, dropout: float = 0)

Bases: torch.nn.Module

Deconvolution block with ConvTranspose2d, Conv2d, batch normalization, ReLU, and dropout.

Parameters:

• in_channels – Number of input channels for deconv block.

• out_channels – Number of output channels for deconv and convolution.

• kernel_size – Kernel size for convolution. Defaults to 3.

• dropout – Dropout rate. Defaults to 0.

block

forward(x)