Class DocumentSplitting

Namespace: LMKit.Extraction

Assembly: LM-Kit.NET.dll

Provides functionality to detect logical document boundaries within a multi-page file using a vision language model (VLM).

public sealed class DocumentSplitting

Inheritance: object

DocumentSplitting

Inherited Members: object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.ReferenceEquals(object, object)

object.ToString()

Examples

Example: Detect documents in a multi-page PDF

using LMKit.Model;
using LMKit.Extraction;
using LMKit.Data;
using System;

// Load a vision-capable model (8B or larger recommended)
LM model = LM.LoadFromModelID("qwen3-vl:8b");

// Create the splitter
DocumentSplitting splitter = new DocumentSplitting(model);

// Analyze a multi-page PDF
DocumentSplittingResult result = splitter.Split(new Attachment("multi_document_scan.pdf"));

// Display results
Console.WriteLine($"Multiple documents: {result.ContainsMultipleDocuments}");
Console.WriteLine($"Document count: {result.DocumentCount}");
Console.WriteLine($"Confidence: {result.Confidence:P0}");

foreach (DocumentSegment segment in result.Segments)
{
    Console.WriteLine($"  Pages {segment.StartPage}-{segment.EndPage}: {segment.Label} ({segment.PageCount} pages)");
}

Example: Use guidance to improve detection accuracy

using LMKit.Model;
using LMKit.Extraction;
using LMKit.Data;
using System;

// Load a vision-capable model (8B or larger recommended)
LM model = LM.LoadFromModelID("qwen3-vl:8b");

// Provide guidance about the expected document types
DocumentSplitting splitter = new DocumentSplitting(model)
{
    Guidance = "The file contains a mix of invoices and purchase orders."
};

DocumentSplittingResult result = splitter.Split(new Attachment("scanned_batch.pdf"));

foreach (DocumentSegment segment in result.Segments)
{
    Console.WriteLine($"Pages {segment.StartPage}-{segment.EndPage}: {segment.Label}");
}

Remarks

The DocumentSplitting class analyzes a multi-page PDF attachment and determines whether it contains multiple logical documents. For each detected document, it returns the page range for each one.

This class requires a vision-capable language model. The model must have HasVision set to true. Page images are fed directly to the VLM for visual boundary detection, which allows reliable splitting even on scanned documents or documents with complex layouts.

Key Features

Detect whether a multi-page PDF contains multiple logical documents
Identify the page range for each detected document
Optional OCR engine integration for scanned documents
Guidance text to improve detection accuracy

Typical Workflow

Create a DocumentSplitting instance with a vision language model
Optionally configure Guidance or LMKit.Extraction.DocumentSplitting.OcrEngine
Call Split(Attachment, CancellationToken) or SplitAsync(Attachment, CancellationToken) with the PDF attachment to analyze
Access results via DocumentSplittingResult

Constructors

DocumentSplitting(LM): Initializes a new instance of the DocumentSplitting class with the specified vision language model.

Properties

Guidance: Gets or sets semantic guidance for the splitting process.

MaximumContextLength: Gets or sets the maximum context length (in tokens) allowed for the language model during splitting.

Model: Gets the vision language model instance used to drive the document splitting process.

Methods

Split(Attachment, CancellationToken): Detects logical document boundaries synchronously within the specified attachment.