IDocumentUnderstandingModel<T>

Interfaces · AiDotNet.VisionLanguage.Interfaces

Interface for document understanding models that process text-heavy images, documents, charts, and tables.

Implements: IGenerativeVisionLanguageModel<T>

Known implementations: DocPedia<T>, GOTOCR2<T>, MPLUGDocOwl15<T>, MPLUGDocOwl2<T>, MPLUGDocOwl<T>, Surya<T>, TextMonkey<T>, UReader<T>

How It Works

Document understanding models specialize in extracting and reasoning about textual content from document images, including OCR, layout analysis, table extraction, and document QA. Architectures include:

  • LayoutLM: Multimodal pre-training with text, layout, and image
  • Donut/Nougat: OCR-free document understanding via image-to-text generation
  • Pix2Struct: Screenshot parsing pre-training for visual language understanding
  • mPLUG-DocOwl: Modular document understanding with visual abstractor

Properties

Property Summary
bool IsOcrFree { get; } Gets whether this model supports OCR-free document understanding (no external OCR needed).

Methods

Method Summary
Tensor<T> AnswerDocumentQuestion(Tensor<T> documentImage, string question) Answers a question about a document image.
Tensor<T> ExtractText(Tensor<T> documentImage) Extracts text content from a document image.