LayoutLMv2<T>
Models & Types · AiDotNet.Document.LayoutAware
LayoutLMv2 neural network for document understanding with visual features.
Inheritance: NeuralNetworkBase<T> → DocumentNeuralNetworkBase<T> → LayoutLMv2<T>
Implements: IDocumentQA<T>, ILayoutDetector<T>
For Beginners
LayoutLMv2 improves on v1 by also looking at the actual image:
- Text content (what the words say)
- Layout structure (where words are positioned)
- Visual appearance (what the document looks like)
Key improvements over v1:
- Visual backbone (ResNeXt-FPN) for image features
- Spatial-aware self-attention mechanism
- Pre-training on both text-layout and image-text-layout alignment
Example usage:
How It Works
LayoutLMv2 extends LayoutLM by adding visual features from a CNN backbone, enabling the model to understand documents through text, layout, AND image features.
Reference: “LayoutLMv2: Multi-modal Pre-training for Visually-rich Document Understanding” (ACL 2021)
https://arxiv.org/abs/2012.14740
Constructors
| Constructor | Summary |
|---|---|
LayoutLMv2(NeuralNetworkArchitecture<T> architecture, LayoutLMv2Options options, ITokenizer tokenizer, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction) |
Creates a LayoutLMv2 model using native layers for training and inference. |
LayoutLMv2(NeuralNetworkArchitecture<T> architecture, string onnxModelPath, ITokenizer tokenizer, LayoutLMv2Options options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction) |
Creates a LayoutLMv2 model using a pre-trained ONNX model for inference. |
Properties
| Property | Summary |
|---|---|
DocumentType SupportedDocumentTypes { get; } |
|
IReadOnlyList<LayoutElementType> SupportedElementTypes { get; } |
|
bool RequiresOCR { get; } |
|
int ExpectedImageSize { get; } |
Methods
| Method | Summary |
|---|---|
Dictionary<string, DocumentQAResult<T>> ExtractFields(Tensor<T> documentImage, IEnumerable<string> fieldPrompts) |
|
Dictionary<string, Tensor<T>> GetNamedLayerActivations(Tensor<T> input) |
|
DocumentLayoutResult<T> DetectLayout(Tensor<T> documentImage) |
|
DocumentLayoutResult<T> DetectLayout(Tensor<T> documentImage, double confidenceThreshold) |
|
DocumentQAResult<T> AnswerQuestion(Tensor<T> documentImage, string question) |
|
DocumentQAResult<T> AnswerQuestion(Tensor<T> documentImage, string question, int maxAnswerLength, double temperature) |
|
IEnumerable<DocumentQAResult<T>> AnswerQuestions(Tensor<T> documentImage, IEnumerable<string> questions) |
|
LayerInputDomain GetInputDomain(int[] inputShape) |
Generated rank-routed external input-domain contract. |
ModelMetadata<T> GetModelMetadata() |
|
Tensor<T> EncodeDocument(Tensor<T> documentImage) |
|
Tensor<T> EncodeMultimodal(Tensor<T> textTokens, Tensor<T> documentImage) |
Full text+image fusion entry (industry-standard LayoutLMv2): encodes BOTH a token-ID sequence and a document image and fuses them through the multimodal transformer. |
Tensor<T> ForwardForTraining(Tensor<T> input) |
|
string GetModelSummary() |
|
void ValidateInputShape(Tensor<T> documentImage) |