VisualBERT<T>

Models & Types · AiDotNet.VisionLanguage.Foundational

VisualBERT: single-stream transformer that concatenates visual and text tokens.

Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → VisualBERT<T>

Implements: IVisionLanguageFusionModel<T>

For Beginners

VisualBERT is a vision-language model. Default values follow the original paper settings.

How It Works

VisualBERT (Li et al., 2019) concatenates Faster R-CNN region features with text token embeddings and processes them in a single BERT-style transformer, enabling implicit cross-modal alignment through self-attention over the combined sequence.

References:

  • Paper: “VisualBERT: A Simple and Performant Baseline for Vision and Language” (Li et al., 2019)

Example

// Create a VisualBERT model for vision-language understanding
// with concatenated visual and text tokens in a single BERT stream
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.TwoDimensional,
    taskType: NeuralNetworkTaskType.ImageClassification,
    inputHeight: 224, inputWidth: 224, inputDepth: 3, outputSize: 512);
            
// ONNX inference mode with pre-trained model
var model = new VisualBERT<double>(architecture, "visualbert.onnx");
            
// Training mode with native layers
var trainModel = new VisualBERT<double>(architecture, new VisualBERTOptions());

Constructors

Constructor Summary
VisualBERT(NeuralNetworkArchitecture<T> architecture, VisualBERTOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
VisualBERT(NeuralNetworkArchitecture<T> architecture, string modelPath, VisualBERTOptions options)

Properties

Property Summary
int EmbeddingDimension { get; }
int FusionEmbeddingDim { get; }
int MaxSequenceLength { get; }

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()
T ComputeMatchingScore(Tensor<T> image, string text)
Tensor<T> EncodeImage(Tensor<T> image)
Tensor<T> ForwardForTraining(Tensor<T> input)
Tensor<T> FuseImageText(Tensor<T> image, string text)