skainet-lang-core/sk.ainet.lang.nn/DilatedConv2d

DilatedConv2d

class DilatedConv2d<T : DType, V>(val inChannels: Int, val outChannels: Int, val kernelSize: Pair<Int, Int>, val dilation: Pair<Int, Int>, val stride: Pair<Int, Int> = 1 to 1, val padding: Pair<Int, Int> = 0 to 0, val groups: Int = 1, val bias: Boolean = true, val name: String = "DilatedConv2d", initWeights: Tensor<T, V>, initBias: Tensor<T, V>? = null) : Module<T, V> , ModuleParameters<T, V> (source)

Dilated (Atrous) Convolution layer.

Dilated convolution introduces gaps (holes) between the kernel elements, effectively increasing the receptive field without increasing the number of parameters or computational cost. This is particularly useful for semantic segmentation and other tasks where capturing multi-scale context is important.

The dilation parameter controls the spacing between kernel elements:

dilation = (1, 1): standard convolution
dilation = (2, 2): skip every other pixel
dilation = (4, 4): skip every 4th pixel

Dilated convolution is also known as "atrous convolution" (from the French word "à trous" meaning "with holes").

This is essentially a convenience wrapper around Conv2d with explicit dilation handling and additional utility methods for dilated convolution operations.

Parameters

inChannels

Number of input channels

outChannels

Number of output channels/filters

kernelSize

Size of the convolving kernel (height, width)

dilation

Spacing between kernel elements (dilation rate)

stride

Stride of the convolution (default: 1, 1)

padding

Padding added to all sides of the input (default: 0, 0)

groups

Number of groups for grouped dilated convolution (default: 1)

bias

Whether to add a learnable bias to the output (default: true)

name

Name of the module

initWeights

Initial weights tensor

initBias

Initial bias tensor (if bias is true)

Constructors

DilatedConv2d

constructor(inChannels: Int, outChannels: Int, kernelSize: Pair<Int, Int>, dilation: Pair<Int, Int>, stride: Pair<Int, Int> = 1 to 1, padding: Pair<Int, Int> = 0 to 0, groups: Int = 1, bias: Boolean = true, name: String = "DilatedConv2d", initWeights: Tensor<T, V>, initBias: Tensor<T, V>? = null)