import type { InferenceSession } from "onnxruntime-common";
import type { CoreCanvas, PlatformProvider } from "../core/platform.js";
import { Base } from "../global.interface.js";
/**
 * Configuration options for face detection models
 */
export interface DetectionModelOptions {
    /** Threshold values for detection */
    threshold: {
        /** Minimum confidence score for face detection */
        confidence: number;
        /** IoU threshold for non-maximum suppression */
        nonMaximumSuppression: number;
    };
    /** Top-K filtering options */
    topK: {
        /** Maximum detections before NMS */
        preNonMaximumSuppression: number;
        /** Maximum detections after NMS */
        postNonMaxiumSuppression: number;
    };
    /** Model input size */
    size: {
        /** Input dimensions [height, width] */
        input: [number, number];
    };
}
/**
 * Options for detect method
 */
export interface DetectOptions {
    /** Threshold values for detection */
    threshold?: {
        /** Minimum confidence score for face detection (default: 0.7) */
        confidence?: number;
        /** IoU threshold for non-maximum suppression (default: 0.4) */
        nonMaximumSuppression?: number;
    };
}
/**
 * Face detection result
 */
export interface DetectionResult {
    /** Bounding box coordinates and dimensions */
    box: {
        x: number;
        y: number;
        width: number;
        height: number;
    };
    /** Confidence score of the detection */
    confidence: number;
    /** Facial landmarks (5 points: left eye, right eye, nose, left mouth, right mouth) */
    landmarks: number[][];
    /** Whether multiple faces were detected in the image */
    multipleFaces: boolean;
}
/**
 * Base class for face detection models
 */
export declare abstract class BaseDetection extends Base {
    /** Class name for logging */
    protected abstract className: string;
    /** Path to the model file */
    protected abstract modelPath: string;
    /** Detection configuration options */
    protected abstract detectionOptions: DetectionModelOptions;
    /** ONNX inference session */
    protected abstract session: InferenceSession | null;
    constructor(platform?: PlatformProvider);
    /** Initializes the detection model */
    abstract initialize(): Promise<void>;
    /**
     * Detects faces in an image
     * @param image - Input image as ArrayBuffer or Canvas
     * @param options - Optional detection options
     * @returns Detection result or null if no face detected
     */
    abstract detect(image: ArrayBuffer | CoreCanvas, options?: DetectOptions): Promise<DetectionResult | null>;
    /**
     * Preprocesses image for model inference
     * @param image - Input canvas
     * @param height - Image height
     * @param width - Image width
     * @returns Preprocessed tensor
     */
    abstract preprocess(image: CoreCanvas, height: number, width: number): Float32Array;
    /**
     * Runs model inference
     * @param tensor - Preprocessed input tensor
     * @param shape - Input shape [height, width]
     * @returns Model outputs
     */
    abstract inference(tensor: Float32Array, shape: [number, number]): Promise<InferenceSession.OnnxValueMapType>;
    /**
     * Postprocesses model outputs
     * @param outputs - Raw model outputs
     * @returns Processed boxes, scores, and landmarks
     */
    abstract postprocess(outputs: InferenceSession.OnnxValueMapType): {
        boxes: Float32Array;
        scores: Float32Array;
        landmarks: Float32Array;
    };
    /** Releases model resources */
    abstract destroy(): Promise<void>;
    /**
     * Checks if the model has been initialized
     * @returns True if initialized
     */
    protected isInitialized(): boolean;
    /**
     * Generates anchor boxes for face detection
     * @param imageSize - Image dimensions [height, width]
     * @returns Flat array of anchor boxes [cx, cy, w, h, ...]
     */
    protected generateAnchors(imageSize?: [number, number]): Float32Array;
    /**
     * Decode locations from predictions using priors to undo
     * the encoding done for offset regression at train time.
     */
    protected decodeBoxes(loc: Float32Array, priors: Float32Array, variances?: [number, number]): Float32Array;
    /**
     * Decode landmark predictions using prior boxes.
     */
    protected decodeLandmarks(predictions: Float32Array, priors: Float32Array, variances?: [number, number]): Float32Array;
    /**
     * Apply Non-Maximum Suppression (NMS) to reduce overlapping bounding boxes.
     */
    protected nonMaximumSuppression(dets: Float32Array, numDets: number, threshold: number): number[];
}