import type { InferenceSession } from "onnxruntime-common";
import type { CoreCanvas, PlatformProvider } from "../core/platform.js";
import { BaseDetection, type DetectOptions, type DetectionModelOptions, type DetectionResult } from "./base.interface.js";
export declare class RetinaNetDetection extends BaseDetection {
    protected className: string;
    protected modelPath: string;
    protected session: InferenceSession | null;
    protected anchorsCache: Float32Array | null;
    protected anchorsCacheShape: string;
    protected detectionOptions: DetectionModelOptions;
    constructor(options?: Partial<DetectionModelOptions>, platform?: PlatformProvider);
    initialize(): Promise<void>;
    /**
     * Detect face in an image, prioritize face that is largest
     * @param image
     * @param options - Optional detection options
     */
    detect(image: ArrayBuffer | CoreCanvas, options?: DetectOptions): Promise<DetectionResult | null>;
    preprocess(canvas: CoreCanvas, height: number, width: number): Float32Array;
    inference(tensor: Float32Array, shape: [number, number]): Promise<InferenceSession.OnnxValueMapType>;
    postprocess(outputs: InferenceSession.OnnxValueMapType): {
        boxes: Float32Array;
        scores: Float32Array;
        landmarks: Float32Array;
    };
    protected postprocessWithThreshold(outputs: InferenceSession.OnnxValueMapType, threshold: {
        confidence: number;
        nonMaximumSuppression: number;
    }): {
        boxes: Float32Array;
        scores: Float32Array;
        landmarks: Float32Array;
    };
    protected applyNMSAndTopK(filteredBoxesArray: Float32Array, filteredScoresArray: Float32Array, filteredLandmarksArray: Float32Array, nmsThreshold: number): {
        boxes: Float32Array;
        scores: Float32Array;
        landmarks: Float32Array;
    };
    destroy(): Promise<void>;
}