import { Metadata as ImageMetadata } from './gtm-image';
import { TrainingParameters as ImageTrainingParams } from './gtm-image/teachable-mobilenet';
import * as tf from '@tensorflow/tfjs';
import { TeachableModel, ExplainedPredictionsOutput, TMType } from './TeachableModel';
import { ExposedMobileNet } from './gtm-image/exposed-mobilenet';
interface TrainingParameters extends ImageTrainingParams {
}
interface BaseMetadata {
    modelBaseUrl?: string;
}
export declare type Metadata = BaseMetadata & ImageMetadata;
export default class ImageModel implements TeachableModel {
    protected model?: ExposedMobileNet;
    protected _ready?: Promise<boolean>;
    protected trained: boolean;
    protected busy: boolean;
    protected imageSize: number;
    protected _disposed: boolean;
    variant: TMType;
    explained?: HTMLCanvasElement;
    modelBaseUrl: string;
    private transferLearningExplainer;
    private readonly trainingLayerName;
    constructor(type: TMType, metadata?: Metadata, model?: tf.io.ModelJSON, weights?: ArrayBuffer);
    getVariant(): TMType;
    estimate(image: HTMLCanvasElement): Promise<HTMLCanvasElement>;
    setXAICanvas(canvas: HTMLCanvasElement): void;
    protected load(metadata?: ImageMetadata, model?: tf.io.ModelJSON, weights?: ArrayBuffer): Promise<void>;
    ready(): Promise<boolean>;
    predict(image: HTMLCanvasElement): Promise<ExplainedPredictionsOutput>;
    train(params: TrainingParameters, callbacks: tf.CustomCallbackArgs): Promise<tf.LayersModel>;
    /**
     * Adds one training sample to a class.
     * Extracts the backbone embedding once from the already-loaded TeachableMobileNet
     * and accumulates it into the per-class ImageNet stats used for XAI explanation,
     * avoiding a redundant full-model forward pass.
     */
    addExample(className: number, image: HTMLCanvasElement): Promise<void>;
    dispose(): void;
    setName(name: string): void;
    getModel(): ExposedMobileNet | undefined;
    getImageSize(): number;
    isTrained(): boolean;
    isReady(): boolean;
    setSeed(seed: string): void;
    getMetadata(): ImageMetadata;
    save(handler: tf.io.IOHandler): Promise<tf.io.SaveResult>;
    /**
     * If a pose is available, draw the keypoints and skeleton.
     *
     * @param image Image to draw the pose into.
     */
    draw(image: HTMLCanvasElement): HTMLCanvasElement;
    setLabels(labels: string[]): void;
    getLabels(): string[];
    getLabel(ix: number): string;
    getNumExamples(): number;
    getExamplesPerClass(): number[];
    getNumValidation(): number;
    calculateAccuracy(): Promise<{
        reference: any;
        predictions: any;
    }>;
}
export {};