EAST<T>

Models & Types · AiDotNet.ComputerVision.Detection.TextDetection

EAST (Efficient and Accurate Scene Text) detector.

Inheritance: ModelBase<T, Tensor<T>, Tensor<T>> → VisionTaskModelBase<T> → TextDetectorBase<T> → EAST<T>

For Beginners

EAST is a fast and accurate text detector that directly predicts word or text-line bounding boxes in one forward pass. It outputs a score map showing where text is likely to be, plus geometry (box coordinates or rotated rectangles) for each text region.

How It Works

Key features:

  • Single-shot detection (no region proposals)
  • Supports both axis-aligned and rotated bounding boxes
  • Fast inference suitable for real-time applications
  • Works well for both horizontal and multi-oriented text

Reference: Zhou et al., “EAST: An Efficient and Accurate Scene Text Detector”, CVPR 2017

Constructors

Constructor Summary
EAST(TextDetectionOptions<T> options, bool useRotatedBoxes) Creates a new EAST text detector: the paper’s VGG16 stem, feature-merging branch and output layer.

Properties

Property Summary
double BoxScoreThreshold { get; set; } The minimum mean score-map value inside a detected box (after locality-aware NMS) for the box to be kept; the per-pixel threshold is the detection call’s confidence threshold.
string Name { get; }

Methods

Method Summary
Task LoadWeightsAsync(string pathOrUrl, CancellationToken cancellationToken)
void SaveWeights(string path)
void SetTrainingMode(bool training)