SVTRMixingBlockLayer<T>
Layers · AiDotNet.NeuralNetworks.Layers
SVTR local/global component-mixing block with the released model’s pre-normalization and MLP.
Inheritance: LayerBase<T> → SVTRMixingBlockLayer<T>
Implements: IShapeContract
For Beginners
Local mixing lets each visual token attend to nearby text strokes, while global mixing lets later blocks connect the full word. The residual MLP then refines each token without changing the image-token grid.
How It Works
Reference: Du et al., “SVTR: Scene Text Recognition with a Single Visual Model”, IJCAI 2022.
Constructors
| Constructor | Summary |
|---|---|
SVTRMixingBlockLayer(int hiddenSize, int numHeads, int height, int width, int windowHeight, int windowWidth, bool local, double dropPathRate) |
Properties
| Property | Summary |
|---|---|
bool SupportsTraining { get; } |
|
bool UsesLocalMixing { get; } |
|
bool UsesPreNormalization { get; } |
|
double DropPathRate { get; } |
|
int GridHeight { get; } |
|
int GridWidth { get; } |
|
int HiddenSize { get; } |
|
int NumHeads { get; } |
|
int WindowHeight { get; } |
|
int WindowWidth { get; } |
Methods
| Method | Summary |
|---|---|
IReadOnlyList<ILayer<T>> GetSubLayers() |
|
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank) |
Derived from this type’s [TensorLayout] declarations. |
void ClearGradients() |
|
void ResetState() |
|
void UpdateParameters(T learningRate) |