VisualBERTOptions
Options & Configuration · AiDotNet.VisionLanguage.Foundational
Configuration options for VisualBERT single-stream fusion model.
Inheritance: ModelOptions → FoundationalVLMOptions → VisualBERTOptions
How It Works
VisualBERT (Li et al., 2019) concatenates visual tokens (from Faster R-CNN) with text tokens in a single BERT transformer stream, allowing implicit cross-modal alignment through self-attention.
Constructors
| Constructor | Summary |
|---|---|
VisualBERTOptions() |
Initializes a new instance with default values. |
VisualBERTOptions(VisualBERTOptions other) |
Initializes a new instance by copying from another instance. |
Properties
| Property | Summary |
|---|---|
bool UseVisualSegmentEmbeddings { get; set; } |
Gets or sets whether to use visual segment embeddings. |
int MaxVisualTokens { get; set; } |
Gets or sets the maximum number of visual tokens. |