TextPolygonTarget

Models & Types · AiDotNet.ComputerVision.Detection.TextDetection

One annotated text region: a polygon in source-image pixels and, optionally, its transcription.

How It Works

Word-level polygons are the standard annotation (ICDAR, Total-Text). CRAFT’s weak supervision uses the transcription’s length to split a word into character boxes (Baek et al. 2019, Section 3.2). Without a transcription it falls back to the word’s aspect ratio.

Constructors

Constructor Summary
TextPolygonTarget(IEnumerable<ValueTuple<double, double>> points, string transcription) Creates a target from at least three polygon vertices, in order around the region.

Properties

Property Summary
IReadOnlyList<ValueTuple<double, double>> Points { get; } The vertices, in source-image pixels.
string Transcription { get; } The region’s text, when known.

Methods

Method Summary
TextPolygonTarget FromBox(double x0, double y0, double x1, double y1, string transcription) An axis-aligned box (x0, y0)..(x1, y1) as a four-vertex polygon.