ViTAdapterOptions
Options & Configuration · AiDotNet.ComputerVision.Segmentation.Semantic
Configuration options for the ViT-Adapter semantic segmentation model.
Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → SegmentationModelOptions → ViTAdapterOptions
For Beginners
ViT-Adapter options inherit from NeuralNetworkOptions, which provides a Seed property for reproducibility. ViT-Adapter enables plain Vision Transformers to handle dense prediction tasks by adding lightweight spatial prior modules, without requiring any vision-specific architectural changes to the base ViT.
Constructors
| Constructor | Summary |
|---|---|
ViTAdapterOptions() |
Initializes a new instance with default values. |
ViTAdapterOptions(ViTAdapterOptions other) |
Initializes a new instance by copying from another instance. |
Properties
| Property | Summary |
|---|---|
ViTAdapterModelSize ModelSize { get; set; } |
Gets or sets which published size variant of the model to build. |
double LearningRate { get; set; } |
Gets or sets the AdamW backbone learning rate used by the ADE20K recipe. |
double WeightDecay { get; set; } |
Gets or sets the AdamW decoupled weight decay used by the ADE20K recipe. |
Methods
| Method | Summary |
|---|---|
void Validate() |
Throws when a value on this instance cannot produce a working model. |