kalidokit

When using Kalidokit solvers, you can provide configuration via ISolveOptions. This interface defines the base settings for both face and pose solvers.

Key properties:

• runtime: Specifies the underlying engine. Defaults to `

If you do not need a full body solve, PoseSolver exposes individual calculators as static methods:

• PoseSolver.calcArms(lm3d): Calculates arm rotations.
• PoseSolver.calcHips(lm3d, lm2d): Calculates hip position and spine rotation.
• PoseSolver.calcLegs(lm3d): Calculates leg rotations.

Note that calcHips requires both 3D and 2D landmarks, while calcArms and calcLegs primarily use 3D landmarks.

The HandSolver.solve() method calculates Euler rotations for the wrist and each finger segment (Proximal, Intermediate, and Distal) based on 3D hand landmarks. It accepts an array of 3D vectors (typically from MediaPipe or TensorFlow.js) and a side parameter to specify if the hand is LEFT or RIGHT.

Returns an object containing the rotation values for the wrist and all finger joints, or undefined if no landmarks are provided.

The PoseSolver.solve() method is the primary entry point for calculating full body rotations and positions. It combines arm, hip, and leg calculations into a single result. It requires both 3D world pose landmarks and 2D pose landmarks.

Parameters

• lm3d: An array of 3D pose vectors (TFVectorPose).
• lm2d: An array of 2D pose vectors (Omit<TFVectorPose, "z">).
• options: An optional configuration object (Partial<IPoseSolveOptions>) containing:
  • runtime: Either `

When calling PoseSolver.solve(), you can provide options to handle different landmark sources and automatic image scaling:

• runtime: Set to "tfjs" or "mediapipe". If set to "tfjs", the solver will normalize the 2D landmarks using the provided imageSize (x/width, y/height).
• video: Can be an HTMLVideoElement or a string selector (e.g., "#my-video"). If provided, the solver automatically extracts the videoWidth and videoHeight to use as imageSize.
• imageSize: An object { width, height } used for normalizing landmarks if the runtime is "tfjs" and video is not provided.

Kalidokit uses several standard types for spatial data:

• XYZ: A record of { x: number, y: number, z: number }.
• LR<T>: A record mapping left and right sides to a type T (e.g., LR<Vector>).
• RotationOrder: Valid strings are "XYZ" | "YZX" | "ZXY" | "XZY" | "YXZ" | "ZYX".
• EulerRotation: An XYZ object that can optionally include a rotationOrder.
• Side: Represents LEFT or RIGHT orientation.

The PoseSolver.solve() method returns a TPose object (or undefined if landmarks are missing). The object contains the following keys:

• RightUpperArm
• RightLowerArm
• LeftUpperArm
• LeftLowerArm
• RightHand
• LeftHand
• RightUpperLeg (Defaults to RestingDefault.Pose.RightUpperLeg if enableLegs is false)
• RightLowerLeg (Defaults to RestingDefault.Pose.RightLowerLeg if enableLegs is false)
• LeftUpperLeg (Defaults to RestingDefault.Pose.LeftUpperLeg if enableLegs is false)
• LeftLowerLeg (Defaults to RestingDefault.Pose.LeftLowerLeg if enableLegs is false)
• Hips
• Spine

The IFaceSolveOptions interface extends ISolveOptions with specific settings for facial tracking.

Key properties:

• smoothBlink: A boolean to toggle blink smoothing. Defaults to false.
• blinkSettings: An array of numbers used to configure blink behavior.

The TPose type represents the processed body pose data. It maps specific body parts to rotation or position data.

Key joints included:

• RightUpperArm, RightLowerArm, LeftUpperArm, LeftLowerArm (Euler rotations)
• RightHand, LeftHand (Vectors)
• RightUpperLeg, RightLowerLeg, LeftUpperLeg, LeftLowerLeg (Euler or XYZ)
• Hips: An IHips object containing position, optional rotation, and optional worldPosition.
• Spine: A Vector or XYZ position.