retinanetjs/build/module/lib/retinanet.d.ts

Wrapper for models built using keras-retinanet.

import * as tf from '@tensorflow/tfjs';
import { AnchorParameters } from './anchors';
/**
 * Represents a detected object with coordinates being provided
 * as percentages of the image width and height.
 */
export interface DetectedObject {
    label: string;
    score: number;
    x1: number;
    x2: number;
    y1: number;
    y2: number;
}
/**
 * Represents a RetinaNet model. Rather than creating directly,
 * it is intended to be created using `load()`.
 */
export declare class RetinaNet {
    readonly model: tf.LayersModel;
    protected readonly classes: string[];
    protected readonly preprocessingMode: string;
    protected readonly anchorParams: AnchorParameters;
    protected readonly height: number;
    protected readonly width: number;
    constructor(model: tf.LayersModel, classes: string[], preprocessingMode: string, anchorParams?: AnchorParameters);
    /**
     * Computes predictions. We currently do not support class-specific filtering.
     * When non-max suppression is applied, it will be across all boxes, regardless of class.
     *
     * @param img The image object on which to run object detection
     * @param threshold The prediction threshold
     * @param nmsThreshold The non-max suppresion IoU threshold
     */
    detect(img: tf.Tensor3D | ImageData | HTMLImageElement | HTMLCanvasElement | HTMLVideoElement, threshold?: number, nmsThreshold?: number): Promise<DetectedObject[]>;
    /**
     * Remove the model from memory.
     */
    dispose(): void;
    private handleImageTensor;
}
/**
 *
 * @param modelPath The path to the model or a `tf.io.IOHandler` object
 * @param classes The list of detected classes
 * @param preprocessingMode One of `tf` or `caffe`. Check the `preprocess_images`
 *   method of your backbone to see which you should use.
 * @param onProgress A callback to report progress
 * @param anchorParams The anchor parameters for your model
 */
export declare function load(modelPath: string | tf.io.IOHandler, classes: string[], preprocessingMode: string, onProgress?: (progress: number, message: string) => void, anchorParams?: AnchorParameters): Promise<RetinaNet>;