@vladmandic/face-api/types/face-api.d.ts

FaceAPI: AI-powered Face Detection & Rotation Tracking, Face Description & Recognition, Age & Gender & Emotion Prediction for Browser and NodeJS using TensorFlow/JS

/// <reference path="../src/types/webgpu.d.ts" />

declare const add: typeof add_;

/**
 * Adds two `tf.Tensor`s element-wise, A + B. Supports broadcasting.
 *
 *
 * ```js
 * const a = tf.tensor1d([1, 2, 3, 4]);
 * const b = tf.tensor1d([10, 20, 30, 40]);
 *
 * a.add(b).print();  // or tf.add(a, b)
 * ```
 *
 * ```js
 * // Broadcast add a with b.
 * const a = tf.scalar(5);
 * const b = tf.tensor1d([10, 20, 30, 40]);
 *
 * a.add(b).print();  // or tf.add(a, b)
 * ```
 * @param a The first `tf.Tensor` to add.
 * @param b The second `tf.Tensor` to add. Must have the same type as `a`.
 *
 * @doc {heading: 'Operations', subheading: 'Arithmetic'}
 */
declare function add_<T extends Tensor>(a: Tensor | TensorLike, b: Tensor | TensorLike): T;

export declare type AgeAndGenderPrediction = {
    age: number;
    gender: Gender;
    genderProbability: number;
};

export declare class AgeGenderNet extends NeuralNetwork<NetParams> {
    private _faceFeatureExtractor;
    constructor(faceFeatureExtractor?: TinyXception);
    get faceFeatureExtractor(): TinyXception;
    runNet(input: NetInput | tf.Tensor4D): NetOutput;
    forwardInput(input: NetInput | tf.Tensor4D): NetOutput;
    forward(input: TNetInput): Promise<NetOutput>;
    predictAgeAndGender(input: TNetInput): Promise<AgeAndGenderPrediction | AgeAndGenderPrediction[]>;
    protected getDefaultModelName(): string;
    dispose(throwOnRedispose?: boolean): void;
    loadClassifierParams(weights: Float32Array): void;
    extractClassifierParams(weights: Float32Array): {
        params: NetParams;
        paramMappings: ParamMapping[];
    };
    protected extractParamsFromWeightMap(weightMap: tf.NamedTensorMap): {
        params: NetParams;
        paramMappings: ParamMapping[];
    };
    protected extractParams(weights: Float32Array): {
        params: NetParams;
        paramMappings: ParamMapping[];
    };
}

export declare const allFaces: typeof allFacesSsdMobilenetv1;

export declare function allFacesSsdMobilenetv1(input: TNetInput, minConfidence?: number): Promise<WithFaceDescriptor<WithFaceLandmarks<WithFaceDetection<{}>>>[]>;

export declare function allFacesTinyYolov2(input: TNetInput, forwardParams?: ITinyYolov2Options): Promise<WithFaceDescriptor<WithFaceLandmarks<WithFaceDetection<{}>>>[]>;

declare enum AnchorPosition {
    TOP_LEFT = "TOP_LEFT",
    TOP_RIGHT = "TOP_RIGHT",
    BOTTOM_LEFT = "BOTTOM_LEFT",
    BOTTOM_RIGHT = "BOTTOM_RIGHT"
}

/** @docalias number[] */
declare interface ArrayMap {
    R0: number;
    R1: number[];
    R2: number[][];
    R3: number[][][];
    R4: number[][][][];
    R5: number[][][][][];
    R6: number[][][][][][];
}

declare const avgPool: typeof avgPool_;

/**
 * Computes the 2D average pooling of an image.
 *
 * @param x The input tensor, of rank 4 or rank 3 of shape
 *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is assumed.
 * @param filterSize The filter size: `[filterHeight, filterWidth]`. If
 *     `filterSize` is a single number, then `filterHeight == filterWidth`.
 * @param strides The strides of the pooling: `[strideHeight, strideWidth]`. If
 *     `strides` is a single number, then `strideHeight == strideWidth`.
 * @param pad The type of padding algorithm:
 *    - `same` and stride 1: output will be of same size as input,
 *       regardless of filter size.
 *    - `valid`: output will be smaller than input if filter is larger
 *       than 1x1.
 *    - For more info, see this guide:
 *     [https://www.tensorflow.org/api_docs/python/tf/nn/convolution](
 *         https://www.tensorflow.org/api_docs/python/tf/nn/convolution)
 * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. If none is
 *     provided, it will default to truncate.
 *
 * @doc {heading: 'Operations', subheading: 'Convolution'}
 */
declare function avgPool_<T extends Tensor3D | Tensor4D>(x: T | TensorLike, filterSize: [number, number] | number, strides: [number, number] | number, pad: 'valid' | 'same' | number | conv_util.ExplicitPadding, dimRoundingMode?: 'floor' | 'round' | 'ceil'): T;

export declare function awaitMediaLoaded(media: HTMLImageElement | HTMLVideoElement | HTMLCanvasElement): Promise<unknown>;

export declare type BatchNorm = {
    sub: tf.Tensor1D;
    truediv: tf.Tensor1D;
};

declare const batchNorm: typeof batchNorm_;

/**
 * Batch normalization.
 *
 * As described in
 * [http://arxiv.org/abs/1502.03167](http://arxiv.org/abs/1502.03167).
 *
 * Mean, variance, scale, and offset can be of two shapes:
 *   - The same shape as the input.
 *   - In the common case, the depth dimension is the last dimension of x, so
 *     the values would be a `tf.Tensor1D` of shape [depth].
 *
 * Also available are stricter rank-specific methods with the same signature
 * as this method that assert that parameters passed are of given rank
 *   - `tf.batchNorm2d`
 *   - `tf.batchNorm3d`
 *   - `tf.batchNorm4d`
 *
 * @param x The input Tensor.
 * @param mean A mean Tensor.
 * @param variance A variance Tensor.
 * @param offset An offset Tensor.
 * @param scale A scale Tensor.
 * @param varianceEpsilon A small float number to avoid dividing by 0.
 *
 * @doc {heading: 'Operations', subheading: 'Normalization'}
 */
declare function batchNorm_<R extends Rank>(x: Tensor<R> | TensorLike, mean: Tensor<R> | Tensor1D | TensorLike, variance: Tensor<R> | Tensor1D | TensorLike, offset?: Tensor<R> | Tensor1D | TensorLike, scale?: Tensor<R> | Tensor1D | TensorLike, varianceEpsilon?: number): Tensor<R>;

export declare class BoundingBox extends Box implements IBoundingBox {
    constructor(left: number, top: number, right: number, bottom: number, allowNegativeDimensions?: boolean);
}

export declare class Box<BoxType = any> implements IBoundingBox, IRect {
    static isRect(rect: any): boolean;
    static assertIsValidBox(box: any, callee: string, allowNegativeDimensions?: boolean): void;
    private _x;
    private _y;
    private _width;
    private _height;
    constructor(_box: IBoundingBox | IRect, allowNegativeDimensions?: boolean);
    get x(): number;
    get y(): number;
    get width(): number;
    get height(): number;
    get left(): number;
    get top(): number;
    get right(): number;
    get bottom(): number;
    get area(): number;
    get topLeft(): Point;
    get topRight(): Point;
    get bottomLeft(): Point;
    get bottomRight(): Point;
    round(): Box<BoxType>;
    floor(): Box<BoxType>;
    toSquare(): Box<BoxType>;
    rescale(s: IDimensions | number): Box<BoxType>;
    pad(padX: number, padY: number): Box<BoxType>;
    clipAtImageBorders(imgWidth: number, imgHeight: number): Box<BoxType>;
    shift(sx: number, sy: number): Box<BoxType>;
    padAtBorders(imageHeight: number, imageWidth: number): {
        dy: number;
        edy: number;
        dx: number;
        edx: number;
        y: number;
        ey: number;
        x: number;
        ex: number;
        w: number;
        h: number;
    };
    calibrate(region: Box): Box<any>;
}

declare type BoxPredictionParams = {
    box_encoding_predictor: ConvParams;
    class_predictor: ConvParams;
};

declare namespace browser {
    export {
        fromPixelsAsync,
        toPixels,
        draw_2 as draw,
        fromPixels
    }
}

/**
 * Creates an IOHandler that loads model artifacts from user-selected files.
 *
 * This method can be used for loading from files such as user-selected files
 * in the browser.
 * When used in conjunction with `tf.loadLayersModel`, an instance of
 * `tf.LayersModel` (Keras-style) can be constructed from the loaded artifacts.
 *
 * ```js
 * // Note: This code snippet won't run properly without the actual file input
 * //   elements in the HTML DOM.
 *
 * // Suppose there are two HTML file input (`<input type="file" ...>`)
 * // elements.
 * const uploadJSONInput = document.getElementById('upload-json');
 * const uploadWeightsInput = document.getElementById('upload-weights');
 * const model = await tf.loadLayersModel(tf.io.browserFiles(
 *     [uploadJSONInput.files[0], uploadWeightsInput.files[0]]));
 * ```
 *
 * @param files `File`s to load from. Currently, this function supports only
 *   loading from files that contain Keras-style models (i.e., `tf.Model`s), for
 *   which an `Array` of `File`s is expected (in that order):
 *   - A JSON file containing the model topology and weight manifest.
 *   - Optionally, one or more binary files containing the binary weights.
 *     These files must have names that match the paths in the `weightsManifest`
 *     contained by the aforementioned JSON file, or errors will be thrown
 *     during loading. These weights files have the same format as the ones
 *     generated by `tensorflowjs_converter` that comes with the `tensorflowjs`
 *     Python PIP package. If no weights files are provided, only the model
 *     topology will be loaded from the JSON file above.
 * @returns An instance of `Files` `IOHandler`.
 *
 * @doc {
 *   heading: 'Models',
 *   subheading: 'Loading',
 *   namespace: 'io',
 *   ignoreCI: true
 * }
 */
declare function browserFiles(files: File[]): IOHandler;

/**
 * Deprecated. Use `tf.io.http`.
 * @param path
 * @param loadOptions
 */
declare function browserHTTPRequest(path: string, loadOptions?: LoadOptions): IOHandler;

export declare function bufferToImage(buf: Blob): Promise<HTMLImageElement>;

declare const cast: typeof cast_;

/**
 * Casts a `tf.Tensor` to a new dtype.
 *
 * ```js
 * const x = tf.tensor1d([1.5, 2.5, 3]);
 * tf.cast(x, 'int32').print();
 * ```
 * @param x The input tensor to be casted.
 * @param dtype The dtype to cast the input tensor to.
 *
 * @doc {heading: 'Tensors', subheading: 'Transformations'}
 */
declare function cast_<T extends Tensor>(x: T | TensorLike, dtype: DataType): T;

/**
 * Check validity of pad when using dimRoundingMode.
 * @param opDesc A string of op description
 * @param pad The type of padding algorithm.
 *   - `same` and stride 1: output will be of same size as input,
 *       regardless of filter size.
 *   - `valid` output will be smaller than input if filter is larger
 *       than 1x1.
 *   - For more info, see this guide:
 *     [https://www.tensorflow.org/api_docs/python/tf/nn/convolution](
 *          https://www.tensorflow.org/api_docs/python/tf/nn/convolution)
 * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. If none is
 *     provided, it will default to truncate.
 * @throws unknown padding parameter
 */
declare function checkPadOnDimRoundingMode(opDesc: string, pad: 'valid' | 'same' | number | ExplicitPadding, dimRoundingMode?: 'floor' | 'round' | 'ceil'): void;

declare const clipByValue: typeof clipByValue_;

/**
 * Clips values element-wise. `max(min(x, clipValueMax), clipValueMin)`
 *
 * ```js
 * const x = tf.tensor1d([-1, 2, -3, 4]);
 *
 * x.clipByValue(-2, 3).print();  // or tf.clipByValue(x, -2, 3)
 * ```
 * @param x The input tensor.
 * @param clipValueMin Lower bound of range to be clipped to.
 * @param clipValueMax Upper bound of range to be clipped to.
 *
 * @doc {heading: 'Operations', subheading: 'Basic math'}
 */
declare function clipByValue_<T extends Tensor>(x: T | TensorLike, clipValueMin: number, clipValueMax: number): T;

export declare class ComposableTask<T> {
    then(onfulfilled: (value: T) => T | PromiseLike<T>): Promise<T>;
    run(): Promise<T>;
}

/**
 * Wraps a list of ArrayBuffers into a `slice()`-able object without allocating
 * a large ArrayBuffer.
 *
 * Allocating large ArrayBuffers (~2GB) can be unstable on Chrome. TFJS loads
 * its weights as a list of (usually) 4MB ArrayBuffers and then slices the
 * weight tensors out of them. For small models, it's safe to concatenate all
 * the weight buffers into a single ArrayBuffer and then slice the weight
 * tensors out of it, but for large models, a different approach is needed.
 */
declare class CompositeArrayBuffer {
    private shards;
    private previousShardIndex;
    private bufferUniformSize?;
    readonly byteLength: number;
    /**
     * Concatenate a number of ArrayBuffers into one.
     *
     * @param buffers An array of ArrayBuffers to concatenate, or a single
     *     ArrayBuffer.
     * @returns Result of concatenating `buffers` in order.
     */
    static join(buffers?: ArrayBuffer[] | ArrayBuffer): ArrayBuffer;
    constructor(buffers?: ArrayBuffer | ArrayBuffer[] | TypedArray | TypedArray[]);
    slice(start?: number, end?: number): ArrayBuffer;
    /**
     * Get the index of the shard that contains the byte at `byteIndex`.
     */
    private findShardForByte;
}

export declare class ComputeAllFaceDescriptorsTask<TSource extends WithFaceLandmarks<WithFaceDetection<{}>>> extends ComputeFaceDescriptorsTaskBase<WithFaceDescriptor<TSource>[], TSource[]> {
    run(): Promise<WithFaceDescriptor<TSource>[]>;
    withFaceExpressions(): PredictAllFaceExpressionsWithFaceAlignmentTask<WithFaceDescriptor<TSource>>;
    withAgeAndGender(): PredictAllAgeAndGenderWithFaceAlignmentTask<WithFaceDescriptor<TSource>>;
}

/**
 * Computes the information for a forward pass of a convolution/pooling
 * operation.
 */
declare function computeConv2DInfo(inShape: [number, number, number, number], filterShape: [number, number, number, number], strides: number | [number, number], dilations: number | [number, number], pad: 'same' | 'valid' | number | ExplicitPadding, roundingMode?: 'floor' | 'round' | 'ceil', depthwise?: boolean, dataFormat?: 'channelsFirst' | 'channelsLast'): Conv2DInfo;

/**
 * Computes the information for a forward pass of a 3D convolution/pooling
 * operation.
 */
declare function computeConv3DInfo(inShape: [number, number, number, number, number], filterShape: [number, number, number, number, number], strides: number | [number, number, number], dilations: number | [number, number, number], pad: 'same' | 'valid' | number, depthwise?: boolean, dataFormat?: 'channelsFirst' | 'channelsLast', roundingMode?: 'floor' | 'round' | 'ceil'): Conv3DInfo;

declare function computeDefaultPad(inputShape: [number, number] | [number, number, number, number], fieldSize: number, stride: number, dilation?: number): number;

/**
 *
 * @param inputShape Input tensor shape is of the following dimensions:
 *     `[batch, height, width, inChannels]`.
 * @param filterShape The filter shape is of the following dimensions:
 *     `[filterHeight, filterWidth, depth]`.
 * @param strides The strides of the sliding window for each dimension of the
 *     input tensor: `[strideHeight, strideWidth]`.
 *     If `strides` is a single number,
 *     then `strideHeight == strideWidth`.
 * @param pad The type of padding algorithm.
 *    - `same` and stride 1: output will be of same size as input,
 *       regardless of filter size.
 *    - `valid`: output will be smaller than input if filter is larger
 *       than 1*1x1.
 *    - For more info, see this guide:
 *     [https://www.tensorflow.org/api_docs/python/tf/nn/convolution](
 *          https://www.tensorflow.org/api_docs/python/tf/nn/convolution)
 * @param dataFormat The data format of the input and output data.
 *     Defaults to 'NHWC'.
 * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`.
 *     Defaults to `[1, 1]`. If `dilations` is a single number, then
 *     `dilationHeight == dilationWidth`.
 */
declare function computeDilation2DInfo(inputShape: [number, number, number, number], filterShape: [number, number, number], strides: number | [number, number], pad: 'same' | 'valid' | number, dataFormat: 'NHWC', dilations: number | [number, number]): Conv2DInfo;

/**
 * Computes a 128 entry vector (face descriptor / face embeddings) from the face shown in an image,
 * which uniquely represents the features of that persons face. The computed face descriptor can
 * be used to measure the similarity between faces, by computing the euclidean distance of two
 * face descriptors.
 *
 * @param inputs The face image extracted from the aligned bounding box of a face. Can
 * also be an array of input images, which will be batch processed.
 * @returns Face descriptor with 128 entries or array thereof in case of batch input.
 */
export declare const computeFaceDescriptor: (input: TNetInput) => Promise<Float32Array | Float32Array[]>;

export declare class ComputeFaceDescriptorsTaskBase<TReturn, TParentReturn> extends ComposableTask<TReturn> {
    protected parentTask: ComposableTask<TParentReturn> | Promise<TParentReturn>;
    protected input: TNetInput;
    constructor(parentTask: ComposableTask<TParentReturn> | Promise<TParentReturn>, input: TNetInput);
}

declare function computePool2DInfo(inShape: [number, number, number, number], filterSize: [number, number] | number, strides: number | [number, number], dilations: number | [number, number], pad: 'same' | 'valid' | number | ExplicitPadding, roundingMode?: 'floor' | 'round' | 'ceil', dataFormat?: 'channelsFirst' | 'channelsLast'): Conv2DInfo;

/**
 * Computes the information for a forward pass of a pooling3D operation.
 */
declare function computePool3DInfo(inShape: [number, number, number, number, number], filterSize: number | [number, number, number], strides: number | [number, number, number], dilations: number | [number, number, number], pad: 'same' | 'valid' | number, roundingMode?: 'floor' | 'round' | 'ceil', dataFormat?: 'NDHWC' | 'NCDHW'): Conv3DInfo;

declare function computeReshapedDimensions({ width, height }: IDimensions, inputSize: number): Dimensions;

export declare class ComputeSingleFaceDescriptorTask<TSource extends WithFaceLandmarks<WithFaceDetection<{}>>> extends ComputeFaceDescriptorsTaskBase<WithFaceDescriptor<TSource> | undefined, TSource | undefined> {
    run(): Promise<WithFaceDescriptor<TSource> | undefined>;
    withFaceExpressions(): PredictSingleFaceExpressionsWithFaceAlignmentTask<WithFaceDescriptor<TSource>>;
    withAgeAndGender(): PredictSingleAgeAndGenderWithFaceAlignmentTask<WithFaceDescriptor<TSource>>;
}

declare const concat: typeof concat_;

/**
 * Concatenates a list of `tf.Tensor`s along a given axis.
 *
 * The tensors ranks and types must match, and their sizes must match in all
 * dimensions except `axis`.
 *
 * Also available are stricter rank-specific methods that assert that
 * `tensors` are of the given rank:
 *   - `tf.concat1d`
 *   - `tf.concat2d`
 *   - `tf.concat3d`
 *   - `tf.concat4d`
 *
 * Except `tf.concat1d` (which does not have axis param), all methods have
 * same signature as this method.
 *
 * ```js
 * const a = tf.tensor1d([1, 2]);
 * const b = tf.tensor1d([3, 4]);
 * a.concat(b).print();  // or a.concat(b)
 * ```
 *
 * ```js
 * const a = tf.tensor1d([1, 2]);
 * const b = tf.tensor1d([3, 4]);
 * const c = tf.tensor1d([5, 6]);
 * tf.concat([a, b, c]).print();
 * ```
 *
 * ```js
 * const a = tf.tensor2d([[1, 2], [10, 20]]);
 * const b = tf.tensor2d([[3, 4], [30, 40]]);
 * const axis = 1;
 * tf.concat([a, b], axis).print();
 * ```
 * @param tensors A list of tensors to concatenate.
 * @param axis The axis to concatenate along. Defaults to 0 (the first dim).
 *
 * @doc {heading: 'Tensors', subheading: 'Slicing and Joining'}
 */
declare function concat_<T extends Tensor>(tensors: Array<T | TensorLike>, axis?: number): T;

/**
 * Concatenate a number of ArrayBuffers into one.
 *
 * @param buffers An array of ArrayBuffers to concatenate, or a single
 *     ArrayBuffer.
 * @returns Result of concatenating `buffers` in order.
 *
 * @deprecated Use tf.io.CompositeArrayBuffer.join() instead.
 */
declare function concatenateArrayBuffers(buffers: ArrayBuffer[] | ArrayBuffer): ArrayBuffer;

declare interface ContextOptions {
    /**
     * Optional.  If the canvas has created a context, it would not make effects.
     * If it is not set, it would be variable based on the current backend.
     */
    contextType?: string;
    /**
     * Optional. A WebGLContextAttributes configuration. If the canvas has created
     * a context, it would not make effects.
     */
    contextAttributes?: WebGLContextAttributes;
}

declare const conv2d: typeof conv2d_;

/**
 * Computes a 2D convolution over the input x.
 *
 * @param x The input tensor, of rank 4 or rank 3, of shape
 *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is
 * assumed.
 * @param filter The filter, rank 4, of shape
 *     `[filterHeight, filterWidth, inDepth, outDepth]`.
 * @param strides The strides of the convolution: `[strideHeight,
 * strideWidth]`.
 * @param pad The type of padding algorithm.
 *    - `same` and stride 1: output will be of same size as input,
 *       regardless of filter size.
 *    - `valid`: output will be smaller than input if filter is larger
 *       than 1x1.
 *   - For more info, see this guide:
 *     [https://www.tensorflow.org/api_docs/python/tf/nn/convolution](
 *          https://www.tensorflow.org/api_docs/python/tf/nn/convolution)
 * @param dataFormat: An optional string from: "NHWC", "NCHW". Defaults to
 *     "NHWC". Specify the data format of the input and output data. With the
 *     default format "NHWC", the data is stored in the order of: [batch,
 *     height, width, channels].
 * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`
 *     in which we sample input values across the height and width dimensions
 *     in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single
 *     number, then `dilationHeight == dilationWidth`. If it is greater than
 *     1, then all values of `strides` must be 1.
 * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. If none is
 *     provided, it will default to truncate.
 *
 * @doc {heading: 'Operations', subheading: 'Convolution'}
 */
declare function conv2d_<T extends Tensor3D | Tensor4D>(x: T | TensorLike, filter: Tensor4D | TensorLike, strides: [number, number] | number, pad: 'valid' | 'same' | number | conv_util.ExplicitPadding, dataFormat?: 'NHWC' | 'NCHW', dilations?: [number, number] | number, dimRoundingMode?: 'floor' | 'round' | 'ceil'): T;

/**
 * Information about the forward pass of a convolution/pooling operation.
 * It includes input and output shape, strides, filter size and padding
 * information.
 */
declare type Conv2DInfo = {
    batchSize: number;
    inHeight: number;
    inWidth: number;
    inChannels: number;
    outHeight: number;
    outWidth: number;
    outChannels: number;
    dataFormat: 'channelsFirst' | 'channelsLast';
    strideHeight: number;
    strideWidth: number;
    dilationHeight: number;
    dilationWidth: number;
    filterHeight: number;
    filterWidth: number;
    effectiveFilterHeight: number;
    effectiveFilterWidth: number;
    padInfo: PadInfo;
    inShape: [number, number, number, number];
    outShape: [number, number, number, number];
    filterShape: [number, number, number, number];
};

/**
 * Information about the forward pass of a 3D convolution/pooling operation.
 * It includes input and output shape, strides, filter size and padding
 * information.
 */
declare type Conv3DInfo = {
    batchSize: number;
    inDepth: number;
    inHeight: number;
    inWidth: number;
    inChannels: number;
    outDepth: number;
    outHeight: number;
    outWidth: number;
    outChannels: number;
    dataFormat: 'channelsFirst' | 'channelsLast';
    strideDepth: number;
    strideHeight: number;
    strideWidth: number;
    dilationDepth: number;
    dilationHeight: number;
    dilationWidth: number;
    filterDepth: number;
    filterHeight: number;
    filterWidth: number;
    effectiveFilterDepth: number;
    effectiveFilterHeight: number;
    effectiveFilterWidth: number;
    padInfo: PadInfo3D;
    inShape: [number, number, number, number, number];
    outShape: [number, number, number, number, number];
    filterShape: [number, number, number, number, number];
};

declare namespace conv_util {
    export {
        computeDilation2DInfo,
        computePool2DInfo,
        computePool3DInfo,
        computeConv2DInfo,
        computeConv3DInfo,
        computeDefaultPad,
        tupleValuesAreOne,
        eitherStridesOrDilationsAreOne,
        stridesOrDilationsArePositive,
        convertConv2DDataFormat,
        checkPadOnDimRoundingMode,
        ExplicitPadding,
        PadInfo,
        PadInfo3D,
        Conv2DInfo,
        Conv3DInfo
    }
}

/**
 * Convert Conv2D dataFormat from 'NHWC'|'NCHW' to
 *    'channelsLast'|'channelsFirst'
 * @param dataFormat in 'NHWC'|'NCHW' mode
 * @return dataFormat in 'channelsLast'|'channelsFirst' mode
 * @throws unknown dataFormat
 */
declare function convertConv2DDataFormat(dataFormat: 'NHWC' | 'NCHW'): 'channelsLast' | 'channelsFirst';

declare type ConvLayerParams = {
    conv: ConvParams;
    scale: ScaleLayerParams;
};

declare type ConvParams = {
    filters: tf.Tensor4D;
    bias: tf.Tensor1D;
};

export declare type ConvWithBatchNorm = {
    conv: ConvParams;
    bn: BatchNorm;
};

/**
 * Copy a model from one URL to another.
 *
 * This function supports:
 *
 * 1. Copying within a storage medium, e.g.,
 *    `tf.io.copyModel('localstorage://model-1', 'localstorage://model-2')`
 * 2. Copying between two storage mediums, e.g.,
 *    `tf.io.copyModel('localstorage://model-1', 'indexeddb://model-1')`
 *
 * ```js
 * // First create and save a model.
 * const model = tf.sequential();
 * model.add(tf.layers.dense(
 *     {units: 1, inputShape: [10], activation: 'sigmoid'}));
 * await model.save('localstorage://demo/management/model1');
 *
 * // Then list existing models.
 * console.log(JSON.stringify(await tf.io.listModels()));
 *
 * // Copy the model, from Local Storage to IndexedDB.
 * await tf.io.copyModel(
 *     'localstorage://demo/management/model1',
 *     'indexeddb://demo/management/model1');
 *
 * // List models again.
 * console.log(JSON.stringify(await tf.io.listModels()));
 *
 * // Remove both models.
 * await tf.io.removeModel('localstorage://demo/management/model1');
 * await tf.io.removeModel('indexeddb://demo/management/model1');
 * ```
 *
 * @param sourceURL Source URL of copying.
 * @param destURL Destination URL of copying.
 * @returns ModelArtifactsInfo of the copied model (if and only if copying
 *   is successful).
 * @throws Error if copying fails, e.g., if no model exists at `sourceURL`, or
 *   if `oldPath` and `newPath` are identical.
 *
 * @doc {
 *   heading: 'Models',
 *   subheading: 'Management',
 *   namespace: 'io',
 *   ignoreCI: true
 * }
 */
declare function copyModel(sourceURL: string, destURL: string): Promise<ModelArtifactsInfo>;

declare function createBrowserEnv(): Environment;

export declare function createCanvas({ width, height }: IDimensions): HTMLCanvasElement;

export declare function createCanvasFromMedia(media: HTMLImageElement | HTMLVideoElement | ImageData, dims?: IDimensions): HTMLCanvasElement;

export declare function createFaceDetectionNet(weights: Float32Array): SsdMobilenetv1;

export declare function createFaceRecognitionNet(weights: Float32Array): FaceRecognitionNet;

declare function createFileSystem(fs?: any): FileSystem_2;

declare function createNodejsEnv(): Environment;

export declare function createSsdMobilenetv1(weights: Float32Array): SsdMobilenetv1;

export declare function createTinyFaceDetector(weights: Float32Array): TinyFaceDetector;

export declare function createTinyYolov2(weights: Float32Array, withSeparableConvs?: boolean): TinyYolov2;

/**
 * We wrap data id since we use weak map to avoid memory leaks.
 * Since we have our own memory management, we have a reference counter
 * mapping a tensor to its data, so there is always a pointer (even if that
 * data is otherwise garbage collectable).
 * See https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/
 * Global_Objects/WeakMap
 */
declare type DataId = object;

declare type DataToGPUOptions = DataToGPUWebGLOption;

declare interface DataToGPUWebGLOption {
    customTexShape?: [number, number];
}

/** @docalias 'float32'|'int32'|'bool'|'complex64'|'string' */
declare type DataType = keyof DataTypeMap;

declare interface DataTypeMap {
    float32: Float32Array;
    int32: Int32Array;
    bool: Uint8Array;
    complex64: Float32Array;
    string: string[];
}

/**
 * Decode flat ArrayBuffer as weights.
 *
 * This function does not handle sharding.
 *
 * This function is the reverse of `encodeWeights`.
 *
 * @param weightData A flat ArrayBuffer or an array of ArrayBuffers carrying the
 *   binary values of the tensors concatenated in the order specified in
 *   `specs`.
 * @param specs Specifications of the names, dtypes and shapes of the tensors
 *   whose value are encoded by `buffer`.
 * @return A map from tensor name to tensor value, with the names corresponding
 *   to names in `specs`.
 * @throws Error, if any of the tensors has unsupported dtype.
 */
declare function decodeWeights(weightData: WeightData, specs: WeightsManifestEntry[]): NamedTensorMap;

declare function decodeWeightsStream(weightStream: ReadableStream<ArrayBuffer>, specs: WeightsManifestEntry[]): Promise<NamedTensorMap>;

export declare type DefaultTinyYolov2NetParams = {
    conv0: ConvWithBatchNorm;
    conv1: ConvWithBatchNorm;
    conv2: ConvWithBatchNorm;
    conv3: ConvWithBatchNorm;
    conv4: ConvWithBatchNorm;
    conv5: ConvWithBatchNorm;
    conv6: ConvWithBatchNorm;
    conv7: ConvWithBatchNorm;
    conv8: ConvParams;
};

declare type DenseBlock3Params = {
    conv0: SeparableConvParams | ConvParams;
    conv1: SeparableConvParams;
    conv2: SeparableConvParams;
};

declare type DenseBlock4Params = DenseBlock3Params & {
    conv3: SeparableConvParams;
};

declare const depthwiseConv2d: typeof depthwiseConv2d_;

/**
 * Depthwise 2D convolution.
 *
 * Given a 4D `input` array and a `filter` array of shape
 * `[filterHeight, filterWidth, inChannels, channelMultiplier]` containing
 * `inChannels` convolutional filters of depth 1, this op applies a
 * different filter to each input channel (expanding from 1 channel to
 * `channelMultiplier` channels for each), then concatenates the results
 * together. The output has `inChannels * channelMultiplier` channels.
 *
 * See
 * [https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d](
 *     https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d)
 * for more details.
 *
 * @param x The input tensor, of rank 4 or rank 3, of shape
 *     `[batch, height, width, inChannels]`. If rank 3, batch of 1 is
 * assumed.
 * @param filter The filter tensor, rank 4, of shape
 *     `[filterHeight, filterWidth, inChannels, channelMultiplier]`.
 * @param strides The strides of the convolution: `[strideHeight,
 * strideWidth]`. If strides is a single number, then `strideHeight ==
 * strideWidth`.
 * @param pad The type of padding algorithm.
 *   - `same` and stride 1: output will be of same size as input,
 *       regardless of filter size.
 *   - `valid`: output will be smaller than input if filter is larger
 *       than 1x1.
 *   - For more info, see this guide:
 *     [https://www.tensorflow.org/api_docs/python/tf/nn/convolution](
 *          https://www.tensorflow.org/api_docs/python/tf/nn/convolution)
 * @param dilations The dilation rates: `[dilationHeight, dilationWidth]`
 *     in which we sample input values across the height and width dimensions
 *     in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single
 *     number, then `dilationHeight == dilationWidth`. If it is greater than
 *     1, then all values of `strides` must be 1.
 * @param dataFormat: An optional string from: "NHWC", "NCHW". Defaults to
 *     "NHWC". Specify the data format of the input and output data. With the
 *     default format "NHWC", the data is stored in the order of: [batch,
 *     height, width, channels]. Only "NHWC" is currently supported.
 * @param dimRoundingMode A string from: 'ceil', 'round', 'floor'. If none is
 *     provided, it will default to truncate.
 *
 * @doc {heading: 'Operations', subheading: 'Convolution'}
 */
declare function depthwiseConv2d_<T extends Tensor3D | Tensor4D>(x: T | TensorLike, filter: Tensor4D | TensorLike, strides: [number, number] | number, pad: 'valid' | 'same' | number | conv_util.ExplicitPadding, dataFormat?: 'NHWC' | 'NCHW', dilations?: [number, number] | number, dimRoundingMode?: 'floor' | 'round' | 'ceil'): T;

export declare class DetectAllFaceLandmarksTask<TSource extends WithFaceDetection<{}>> extends DetectFaceLandmarksTaskBase<WithFaceLandmarks<TSource>[], TSource[]> {
    run(): Promise<WithFaceLandmarks<TSource>[]>;
    withFaceExpressions(): PredictAllFaceExpressionsWithFaceAlignmentTask<WithFaceLandmarks<TSource>>;
    withAgeAndGender(): PredictAllAgeAndGenderWithFaceAlignmentTask<WithFaceLandmarks<TSource>>;
    withFaceDescriptors(): ComputeAllFaceDescriptorsTask<WithFaceLandmarks<TSource>>;
}

export declare function detectAllFaces(input: TNetInput, options?: FaceDetectionOptions): DetectAllFacesTask;

export declare class DetectAllFacesTask extends DetectFacesTaskBase<FaceDetection[]> {
    run(): Promise<FaceDetection[]>;
    private runAndExtendWithFaceDetections;
    withFaceLandmarks(useTinyLandmarkNet?: boolean): DetectAllFaceLandmarksTask<{
        detection: FaceDetection;
    }>;
    withFaceExpressions(): PredictAllFaceExpressionsTask<{
        detection: FaceDetection;
    }>;
    withAgeAndGender(): PredictAllAgeAndGenderTask<{
        detection: FaceDetection;
    }>;
}

/**
 * Detects the 68 point face landmark positions of the face shown in an image.
 *
 * @param inputs The face image extracted from the bounding box of a face. Can
 * also be an array of input images, which will be batch processed.
 * @returns 68 point face landmarks or array thereof in case of batch input.
 */
export declare const detectFaceLandmarks: (input: TNetInput) => Promise<FaceLandmarks68 | FaceLandmarks68[]>;

export declare class DetectFaceLandmarksTaskBase<TReturn, TParentReturn> extends ComposableTask<TReturn> {
    protected parentTask: ComposableTask<TParentReturn> | Promise<TParentReturn>;
    protected input: TNetInput;
    protected useTinyLandmarkNet: boolean;
    constructor(parentTask: ComposableTask<TParentReturn> | Promise<TParentReturn>, input: TNetInput, useTinyLandmarkNet: boolean);
    protected get landmarkNet(): FaceLandmark68Net | FaceLandmark68TinyNet;
}

/**
 * Detects the 68 point face landmark positions of the face shown in an image
 * using a tinier version of the 68 point face landmark model, which is slightly
 * faster at inference, but also slightly less accurate.
 *
 * @param inputs The face image extracted from the bounding box of a face. Can
 * also be an array of input images, which will be batch processed.
 * @returns 68 point face landmarks or array thereof in case of batch input.
 */
export declare const detectFaceLandmarksTiny: (input: TNetInput) => Promise<FaceLandmarks68 | FaceLandmarks68[]>;

export declare class DetectFacesTaskBase<TReturn> extends ComposableTask<TReturn> {
    protected input: TNetInput;
    protected options: FaceDetectionOptions;
    constructor(input: TNetInput, options?: FaceDetectionOptions);
}

export declare const detectLandmarks: (input: TNetInput) => Promise<FaceLandmarks68 | FaceLandmarks68[]>;

export declare function detectSingleFace(input: TNetInput, options?: FaceDetectionOptions): DetectSingleFaceTask;

export declare class DetectSingleFaceLandmarksTask<TSource extends WithFaceDetection<{}>> extends DetectFaceLandmarksTaskBase<WithFaceLandmarks<TSource> | undefined, TSource | undefined> {
    run(): Promise<WithFaceLandmarks<TSource> | undefined>;
    withFaceExpressions(): PredictSingleFaceExpressionsWithFaceAlignmentTask<WithFaceLandmarks<TSource>>;
    withAgeAndGender(): PredictSingleAgeAndGenderWithFaceAlignmentTask<WithFaceLandmarks<TSource>>;
    withFaceDescriptor(): ComputeSingleFaceDescriptorTask<WithFaceLandmarks<TSource>>;
}

export declare class DetectSingleFaceTask extends DetectFacesTaskBase<FaceDetection | undefined> {
    run(): Promise<FaceDetection | undefined>;
    private runAndExtendWithFaceDetection;
    withFaceLandmarks(useTinyLandmarkNet?: boolean): DetectSingleFaceLandmarksTask<{
        detection: FaceDetection;
    }>;
    withFaceExpressions(): PredictSingleFaceExpressionsTask<{
        detection: FaceDetection;
    }>;
    withAgeAndGender(): PredictSingleAgeAndGenderTask<{
        detection: FaceDetection;
    }>;
}

export declare class Dimensions implements IDimensions {
    private _width;
    private _height;
    constructor(width: number, height: number);
    get width(): number;
    get height(): number;
    reverse(): Dimensions;
}

declare const div: typeof div_;

/**
 * Divides two `tf.Tensor`s element-wise, A / B. Supports broadcasting.
 *
 * ```js
 * const a = tf.tensor1d([1, 4, 9, 16]);
 * const b = tf.tensor1d([1, 2, 3, 4]);
 *
 * a.div(b).print();  // or tf.div(a, b)
 * ```
 *
 * ```js
 * // Broadcast div a with b.
 * const a = tf.tensor1d([2, 4, 6, 8]);
 * const b = tf.scalar(2);
 *
 * a.div(b).print();  // or tf.div(a, b)
 * ```
 *
 * @param a The first tensor as the numerator.
 * @param b The second tensor as the denominator. Must have the same dtype as
 * `a`.
 *
 * @doc {heading: 'Operations', subheading: 'Arithmetic'}
 */
declare function div_<T extends Tensor>(a: Tensor | TensorLike, b: Tensor | TensorLike): T;

declare namespace draw {
    export {
        drawContour,
        drawDetections,
        TDrawDetectionsInput,
        drawFaceExpressions,
        DrawFaceExpressionsInput,
        IDrawBoxOptions,
        DrawBoxOptions,
        DrawBox,
        drawFaceLandmarks,
        IDrawFaceLandmarksOptions,
        DrawFaceLandmarksOptions,
        DrawFaceLandmarks,
        DrawFaceLandmarksInput,
        AnchorPosition,
        IDrawTextFieldOptions,
        DrawTextFieldOptions,
        DrawTextField
    }
}
export { draw }

/**
 * Draws a `tf.Tensor` to a canvas.
 *
 * When the dtype of the input is 'float32', we assume values in the range
 * [0-1]. Otherwise, when input is 'int32', we assume values in the range
 * [0-255].
 *
 * @param image The tensor to draw on the canvas. Must match one of
 * these shapes:
 *   - Rank-2 with shape `[height, width`]: Drawn as grayscale.
 *   - Rank-3 with shape `[height, width, 1]`: Drawn as grayscale.
 *   - Rank-3 with shape `[height, width, 3]`: Drawn as RGB with alpha set in
 *     `imageOptions` (defaults to 1, which is opaque).
 *   - Rank-3 with shape `[height, width, 4]`: Drawn as RGBA.
 * @param canvas The canvas to draw to.
 * @param options The configuration arguments for image to be drawn and the
 *     canvas to draw to.
 *
 * @doc {heading: 'Browser', namespace: 'browser'}
 */
declare function draw_2(image: Tensor2D | Tensor3D | TensorLike, canvas: HTMLCanvasElement, options?: DrawOptions): void;

declare class DrawBox {
    box: Box;
    options: DrawBoxOptions;
    constructor(box: IBoundingBox | IRect, options?: IDrawBoxOptions);
    draw(canvasArg: string | HTMLCanvasElement | CanvasRenderingContext2D): void;
}

declare class DrawBoxOptions {
    boxColor: string;
    lineWidth: number;
    drawLabelOptions: DrawTextFieldOptions;
    label?: string;
    constructor(options?: IDrawBoxOptions);
}

declare function drawContour(ctx: CanvasRenderingContext2D, points: Point[], isClosed?: boolean): void;

declare function drawDetections(canvasArg: string | HTMLCanvasElement, detections: TDrawDetectionsInput | Array<TDrawDetectionsInput>): void;

declare function drawFaceExpressions(canvasArg: string | HTMLCanvasElement, faceExpressions: DrawFaceExpressionsInput | Array<DrawFaceExpressionsInput>, minConfidence?: number, textFieldAnchor?: IPoint): void;

declare type DrawFaceExpressionsInput = FaceExpressions | WithFaceExpressions<{}>;

declare class DrawFaceLandmarks {
    faceLandmarks: FaceLandmarks;
    options: DrawFaceLandmarksOptions;
    constructor(faceLandmarks: FaceLandmarks, options?: IDrawFaceLandmarksOptions);
    draw(canvasArg: string | HTMLCanvasElement | CanvasRenderingContext2D): void;
}

declare function drawFaceLandmarks(canvasArg: string | HTMLCanvasElement, faceLandmarks: DrawFaceLandmarksInput | Array<DrawFaceLandmarksInput>): void;

declare type DrawFaceLandmarksInput = FaceLandmarks | WithFaceLandmarks<WithFaceDetection<{}>>;

declare class DrawFaceLandmarksOptions {
    drawLines: boolean;
    drawPoints: boolean;
    lineWidth: number;
    pointSize: number;
    lineColor: string;
    pointColor: string;
    constructor(options?: IDrawFaceLandmarksOptions);
}

declare interface DrawOptions {
    /**
     * Optional. An object of options to customize the values of image tensor.
     */
    imageOptions?: ImageOptions;
    /**
     * Optional. An object to configure the context of the canvas to draw to.
     */
    contextOptions?: ContextOptions;
}

declare class DrawTextField {
    text: string[];
    anchor: IPoint;
    options: DrawTextFieldOptions;
    constructor(text: string | string[] | DrawTextField, anchor: IPoint, options?: IDrawTextFieldOptions);
    measureWidth(ctx: CanvasRenderingContext2D): number;
    measureHeight(): number;
    getUpperLeft(ctx: CanvasRenderingContext2D, canvasDims?: IDimensions): IPoint;
    draw(canvasArg: string | HTMLCanvasElement | CanvasRenderingContext2D): void;
}

declare class DrawTextFieldOptions implements IDrawTextFieldOptions {
    anchorPosition: AnchorPosition;
    backgroundColor: string;
    fontColor: string;
    fontSize: number;
    fontStyle: string;
    padding: number;
    constructor(options?: IDrawTextFieldOptions);
}

declare function eitherStridesOrDilationsAreOne(strides: number | number[], dilations: number | number[]): boolean;

/**
 * Encode a map from names to weight values as an ArrayBuffer, along with an
 * `Array` of `WeightsManifestEntry` as specification of the encoded weights.
 *
 * This function does not perform sharding.
 *
 * This function is the reverse of `decodeWeights`.
 *
 * @param tensors A map ("dict") from names to tensors.
 * @param group Group to which the weights belong (optional).
 * @returns A `Promise` of
 *   - A flat `ArrayBuffer` with all the binary values of the `Tensor`s
 *     concatenated.
 *   - An `Array` of `WeightManifestEntry`s, carrying information including
 *     tensor names, `dtype`s and shapes.
 * @throws Error: on unsupported tensor `dtype`.
 */
declare function encodeWeights(tensors: NamedTensorMap | NamedTensor[], group?: WeightGroup): Promise<{
    data: ArrayBuffer;
    specs: WeightsManifestEntry[];
}>;

declare let ENV: Environment_2;

export declare const env: {
    getEnv: typeof getEnv;
    setEnv: typeof setEnv;
    initialize: typeof initialize;
    createBrowserEnv: typeof createBrowserEnv;
    createFileSystem: typeof createFileSystem;
    createNodejsEnv: typeof createNodejsEnv;
    monkeyPatch: typeof monkeyPatch;
    isBrowser: typeof isBrowser;
    isNodejs: typeof isNodejs;
};

export declare type Environment = FileSystem_2 & {
    Canvas: typeof HTMLCanvasElement;
    CanvasRenderingContext2D: typeof CanvasRenderingContext2D;
    Image: typeof HTMLImageElement;
    ImageData: typeof ImageData;
    Video: typeof HTMLVideoElement;
    createCanvasElement: () => HTMLCanvasElement;
    createImageElement: () => HTMLImageElement;
    createVideoElement: () => HTMLVideoElement;
    fetch: (url: string, init?: RequestInit) => Promise<Response>;
};

/**
 * The environment contains evaluated flags as well as the registered platform.
 * This is always used as a global singleton and can be retrieved with
 * `tf.env()`.
 *
 * @doc {heading: 'Environment'}
 */
declare class Environment_2 {
    global: any;
    private flags;
    private flagRegistry;
    private urlFlags;
    platformName: string;
    platform: Platform;
    getQueryParams: typeof getQueryParams;
    constructor(global: any);
    setPlatform(platformName: string, platform: Platform): void;
    registerFlag(flagName: string, evaluationFn: FlagEvaluationFn, setHook?: (value: FlagValue) => void): void;
    getAsync(flagName: string): Promise<FlagValue>;
    get(flagName: string): FlagValue;
    getNumber(flagName: string): number;
    getBool(flagName: string): boolean;
    getString(flagName: string): string;
    getFlags(): Flags;
    get features(): Flags;
    set(flagName: string, value: FlagValue): void;
    private evaluateFlag;
    setFlags(flags: Flags): void;
    reset(): void;
    private populateURLFlags;
}

export declare function euclideanDistance(arr1: number[] | Float32Array, arr2: number[] | Float32Array): number;

declare const exp: typeof exp_;

/**
 * Computes exponential of the input `tf.Tensor` element-wise. `e ^ x`
 *
 * ```js
 * const x = tf.tensor1d([1, 2, -3]);
 *
 * x.exp().print();  // or tf.exp(x)
 * ```
 * @param x The input tensor.
 *
 * @doc {heading: 'Operations', subheading: 'Basic math'}
 */
declare function exp_<T extends Tensor>(x: T | TensorLike): T;

declare const expandDims: typeof expandDims_;

/**
 * Returns a `tf.Tensor` that has expanded rank, by inserting a dimension
 * into the tensor's shape.
 *
 * ```js
 * const x = tf.tensor1d([1, 2, 3, 4]);
 * const axis = 1;
 * x.expandDims(axis).print();
 * ```
 *
 * @param x The input tensor whose dimensions are to be expanded.
 * @param axis The dimension index at which to insert shape of `1`. Defaults
 *     to 0 (the first dimension).
 *
 * @doc {heading: 'Tensors', subheading: 'Transformations'}
 */
declare function expandDims_<T extends Tensor>(x: Tensor | TensorLike, axis?: number): T;

declare type ExplicitPadding = [
[number, number],
[number, number],
[number, number],
[number, number]
];

export declare function extendWithAge<TSource>(sourceObj: TSource, age: number): WithAge<TSource>;

export declare function extendWithFaceDescriptor<TSource>(sourceObj: TSource, descriptor: Float32Array): WithFaceDescriptor<TSource>;

export declare function extendWithFaceDetection<TSource>(sourceObj: TSource, detection: FaceDetection): WithFaceDetection<TSource>;

export declare function extendWithFaceExpressions<TSource>(sourceObj: TSource, expressions: FaceExpressions): WithFaceExpressions<TSource>;

export declare function extendWithFaceLandmarks<TSource extends WithFaceDetection<{}>, TFaceLandmarks extends FaceLandmarks = FaceLandmarks68>(sourceObj: TSource, unshiftedLandmarks: TFaceLandmarks): WithFaceLandmarks<TSource, TFaceLandmarks>;

export declare function extendWithGender<TSource>(sourceObj: TSource, gender: Gender, genderProbability: number): WithGender<TSource>;

/**
 * Extracts the image regions containing the detected faces.
 *
 * @param input The image that face detection has been performed on.
 * @param detections The face detection results or face bounding boxes for that image.
 * @returns The Canvases of the corresponding image region for each detected face.
 */
export declare function extractFaces(input: TNetInput, detections: Array<FaceDetection | Rect>): Promise<HTMLCanvasElement[]>;

/**
 * Extracts the tensors of the image regions containing the detected faces.
 * Useful if you want to compute the face descriptors for the face images.
 * Using this method is faster then extracting a canvas for each face and
 * converting them to tensors individually.
 *
 * @param imageTensor The image tensor that face detection has been performed on.
 * @param detections The face detection results or face bounding boxes for that image.
 * @returns Tensors of the corresponding image region for each detected face.
 */
export declare function extractFaceTensors(imageTensor: tf.Tensor3D | tf.Tensor4D, detections: Array<FaceDetection | Rect>): Promise<tf.Tensor3D[]>;

export declare const FACE_EXPRESSION_LABELS: readonly ["neutral", "happy", "sad", "angry", "fearful", "disgusted", "surprised"];

export declare class FaceDetection extends ObjectDetection implements IFaceDetecion {
    constructor(score: number, relativeBox: Rect, imageDims: IDimensions);
    forSize(width: number, height: number): FaceDetection;
}

export declare type FaceDetectionFunction = (input: TNetInput) => Promise<FaceDetection[]>;

export declare class FaceDetectionNet extends SsdMobilenetv1 {
}

export declare type FaceDetectionOptions = TinyFaceDetectorOptions | SsdMobilenetv1Options | TinyYolov2Options;

export declare class FaceExpressionNet extends FaceProcessor<FaceFeatureExtractorParams> {
    constructor(faceFeatureExtractor?: FaceFeatureExtractor);
    forwardInput(input: NetInput | tf.Tensor4D): tf.Tensor2D;
    forward(input: TNetInput): Promise<tf.Tensor2D>;
    predictExpressions(input: TNetInput): Promise<FaceExpressions | FaceExpressions[]>;
    protected getDefaultModelName(): string;
    protected getClassifierChannelsIn(): number;
    protected getClassifierChannelsOut(): number;
}

export declare class FaceExpressions {
    neutral: number;
    happy: number;
    sad: number;
    angry: number;
    fearful: number;
    disgusted: number;
    surprised: number;
    constructor(probabilities: number[] | Float32Array);
    asSortedArray(): {
        expression: "neutral" | "happy" | "sad" | "angry" | "fearful" | "disgusted" | "surprised";
        probability: number;
    }[];
}

declare class FaceFeatureExtractor extends NeuralNetwork<FaceFeatureExtractorParams> implements IFaceFeatureExtractor<FaceFeatureExtractorParams> {
    constructor();
    forwardInput(input: NetInput): tf.Tensor4D;
    forward(input: TNetInput): Promise<tf.Tensor4D>;
    protected getDefaultModelName(): string;
    protected extractParamsFromWeightMap(weightMap: tf.NamedTensorMap): {
        params: FaceFeatureExtractorParams;
        paramMappings: ParamMapping[];
    };
    protected extractParams(weights: Float32Array): {
        params: FaceFeatureExtractorParams;
        paramMappings: ParamMapping[];
    };
}

declare type FaceFeatureExtractorParams = {
    dense0: DenseBlock4Params;
    dense1: DenseBlock4Params;
    dense2: DenseBlock4Params;
    dense3: DenseBlock4Params;
};

export declare class FaceLandmark68Net extends FaceLandmark68NetBase<FaceFeatureExtractorParams> {
    constructor(faceFeatureExtractor?: FaceFeatureExtractor);
    protected ge