SwinPatchMergingLayer<T>

Layers · AiDotNet.NeuralNetworks.Layers

Patch merging layer for Swin Transformer that performs downsampling between stages.

Inheritance: LayerBase<T> → SwinPatchMergingLayer<T>

Implements: IShapeContract

For Beginners

Think of this like pooling in CNNs, but instead of taking max or average, we concatenate 4 neighboring patches together (2x2 grid) and then use a linear layer to reduce the combined channels. This lets the network process information at multiple scales.

How It Works

This layer merges 2x2 neighboring patches into a single patch, reducing spatial resolution by half while doubling the channel dimension. This creates the hierarchical structure characteristic of Swin Transformer.

Reference: Liu et al., “Swin Transformer: Hierarchical Vision Transformer using Shifted Windows”, ICCV 2021

Constructors

Constructor Summary
SwinPatchMergingLayer(int inputDim) Creates a new Swin patch merging layer.

Properties

Property Summary
bool SupportsTraining { get; }

Methods

Method Summary
IReadOnlyList<ILayer<T>> GetSubLayers()
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank)
void ClearGradients()
void ResetState()
void UpdateParameters(T learningRate)