@handtracking.io/yoha/dist/core/post_model/post_model.js

Yoha is currently available for the web via JavaScript. More languages will be added in the future. If you want to port Yoha to another language and need help feel free reach out.

import { AspectRatioAwareRotation, CoordsOutsideBox, FlipCoordsHorizontally, CoordsInsideBox } from '../../util/math_helper';
const RECOMMENDED_CLASSIFIER_THRESHOLDS = {
    PINCH: 0.2,
    FIST: 0.5,
    IS_HAND_PRESENT: 0.5,
    IS_LEFT_HAND: 0.5,
};
/**
 * @public
 * Recommended thresholds for probabilities reported in {@link IPoseProbabilities},
 * {@link ITrackResult.isLeftHandProb} and {@link ITrackResult.isHandPresentProb}.
 * @privateRemarks
 * These thresholds are an optimization. Math.round(probability) works fine in general.
 * The thresholds have been determined empirically.
 */
export const RecommendedHandPoseProbabilityThresholds = RECOMMENDED_CLASSIFIER_THRESHOLDS;
const USER_FRIENDLY_COORD_ORDER = [
    17,
    16,
    18,
    19,
    1,
    0,
    2,
    3,
    5,
    4,
    6,
    7,
    9,
    8,
    10,
    11,
    13,
    12,
    14,
    15,
    20,
];
/**
 * Converts landmark coordinates (that are with respect to some bounding box) to global
 * coordinates (that are with respect to the original frame).
 *
 * @param preprocInfo - The preprocessing info of the box.
 * @param aspectRatio - [width, height] of the original frame.
 * @param coords - List of landmark coordinates.
 */
export function ConvertLandmarkCoordinatesToGlobalCoordinates(preprocInfo, aspectRatio, coords) {
    const p = preprocInfo;
    const araRot = new AspectRatioAwareRotation(p.rotationCenter, p.rotationInRadians, aspectRatio);
    coords = CoordsOutsideBox([p.topLeft, p.bottomRight], coords);
    coords = araRot.ApplyReverse(coords);
    return coords;
}
/**
 * Converts list of landmark coordinates and classifier probabilities to
 * canonical tracking result structure.
 *
 * @param coords - List of coordinates from landmark model.
 * @param classes - List of classifier probabilities from landmark model.
 */
export function AssembleTrackResultFromCoordinatesAndClasses(coords, classes) {
    return {
        coordinates: coords,
        poses: {
            pinchProb: classes[0],
            fistProb: classes[1],
        },
        isLeftHandProb: classes[3],
        // The class for hand presence is special case as it actually predicts 1
        // if there is no hand present so we need to invert the probability before applying.
        // the threshold.
        isHandPresentProb: (1 - classes[2]),
    };
}
/**
 * @public
 * Mirrors result coordinates along the y axis i.e. `[x, y]` becomes `[1 - x, y]`.
 * @param coords - The coordinates to mirror.
 */
export function MirrorCoordinatesHorizontally(coords) {
    return FlipCoordsHorizontally(coords);
}
/**
 * @public
 * Starting from the borders, removes a percentage of the analyzed track source.
 *
 * @example
 * Example: A video with resolution `640x480` would yield landmark coordinates where
 * `[0, 0]` and `[1, 1]` correspond to pixel values of `[0, 0]` and `[639, 639]` respectively.
 *
 * Setting padding to `0.05` would make `[0, 0]` and `[1, 1]` correspond to pixel values of
 * `[639 * 0.05, 639 * 0.05]` and `[639 - 639 * 0.05, 639 - 639 * 0.05]` respectively.
 *
 * @param padding - The padding to apply.
 * @param coords - The coordinates on which to apply the padding.
 * @returns List of coordinates with the given padding applied.
 */
export function ApplyPaddingToCoordinates(padding, coords) {
    const d = padding;
    const topLeft = [d, d];
    const bottomRight = [1 - d, 1 - d];
    coords = CoordsInsideBox([topLeft, bottomRight], coords);
    return coords;
}
/**
 * Reorders coordinates to have a more intuitive order (as opposed to the one coming from the
 * model).
 */
export function MakeCoordinateOrderUserFriendly(coords) {
    const newCoords = [];
    for (const i of USER_FRIENDLY_COORD_ORDER) {
        newCoords.push(coords[i]);
    }
    return newCoords;
}
//# sourceMappingURL=post_model.js.map