SVTROptions
Options & Configuration · AiDotNet.Document.Options
Configuration options for the SVTR document model.
Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → DocumentNeuralNetworkOptions → SVTROptions
Constructors
| Constructor | Summary |
|---|---|
SVTROptions() |
Creates options with the default values. |
SVTROptions(SVTROptions other) |
Creates a copy of another options instance. |
Properties
| Property | Summary |
|---|---|
bool UseTpsRectification { get; set; } |
Whether the training recipe uses its optional TPS rectifier before SVTRNet. |
double DropPathRate { get; set; } |
Maximum stochastic-depth probability, linearly scheduled across 12 blocks. |
double LastStageDropout { get; set; } |
Dropout applied immediately before the CTC head. |
double TpsMarginX { get; set; } |
Horizontal inset of the target control-point rectangle. |
double TpsMarginY { get; set; } |
Vertical inset of the target control-point rectangle. |
int InputHeight { get; set; } |
Post-TPS input height used by the reference SVTR-Tiny network. |
int InputWidth { get; set; } |
Post-TPS input width used by the reference SVTR-Tiny network. |
int LocalMixingBlocks { get; set; } |
Number of local-mixing blocks before global mixing begins. |
int LocalWindowHeight { get; set; } |
Local component-mixing window height. |
int LocalWindowWidth { get; set; } |
Local component-mixing window width. |
int OutputChannels { get; set; } |
Channel width presented to the CTC head. |
int OutputCharacterPositions { get; } |
Width of the recognition sequence after height collapse: the stem halves the width twice, so InputWidth / 4. |
int TpsControlPointCount { get; set; } |
Even number of boundary control points predicted by the TPS head. |
int TpsInputHeight { get; set; } |
Height of the TPS localization-network input. |
int TpsInputWidth { get; set; } |
Width of the TPS localization-network input. |
int[] EmbedDimensions { get; set; } |
Stage embedding dimensions. |
int[] StageDepths { get; set; } |
Component-mixing block count in each stage. |
int[] StageHeads { get; set; } |
Attention head count in each stage. |
string Charset { get; set; } |
Gets or sets charset. |
Methods
| Method | Summary |
|---|---|
void Validate() |
Throws when a value on this instance cannot produce a working model. |
void ValidateReferenceTopology() |
Validates that the three-stage SVTR-Tiny topology is internally consistent. |