usePoseDetection

Creates a native pose detector and returns a VisionCamera frame processor plus the handlers needed to map results onto the screen.

function usePoseDetection(
  callbacks: DetectionCallbacks<PoseDetectionResultBundle>,
  runningMode: RunningMode,
  model: string,
  options?: Partial<PoseDetectionOptions>
): MediaPipeSolution;

Parameters

callbacks

Field	Type	Description
onResults	(result: PoseDetectionResultBundle, vc: ViewCoordinator) => void	Called with each detection result. vc converts landmarks to view coordinates, see skeleton overlay.
onError	(error: DetectionError) => void	Called when native detection fails. error has code and message.

Wrap both in useCallback so they don’t change on every render.

runningMode

Value	Use for
RunningMode.LIVE_STREAM	Camera frames. This is what you want with VisionCamera.
RunningMode.VIDEO	Decoded video frames processed in order.
RunningMode.IMAGE	Single images. Prefer PoseDetectionOnImage instead.

model

File name of a bundled MediaPipe pose landmarker model, for example 'pose_landmarker_lite.task'. See Model files.

options

All fields are optional.

Option	Type	Default	Description
numPoses	number	1	Maximum number of people to detect.
minPoseDetectionConfidence	number	0.5	Minimum score (0–1) for a person to be detected.
minPosePresenceConfidence	number	0.5	Minimum presence score (0–1) for a detected pose.
minTrackingConfidence	number	0.5	Minimum score (0–1) to keep tracking between frames instead of re-detecting.
shouldOutputSegmentationMasks	boolean	false	Ask MediaPipe for segmentation masks. Slower.
delegate	Delegate	Delegate.GPU	Delegate.GPU or Delegate.CPU.
fpsMode	'none' | number	'none'	Process at most this many frames per second. See Performance.
mirrorMode	'no-mirror' | 'mirror' | 'mirror-front-only'	'mirror-front-only' on Android, 'no-mirror' on iOS	Whether converted points are mirrored horizontally.
forceCameraOrientation	ImageOrientation	—	Override the detected camera sensor orientation.
forceOutputOrientation	ImageOrientation	—	Override the output orientation sent to MediaPipe.

Changing model, runningMode, delegate, numPoses, any confidence threshold or shouldOutputSegmentationMasks releases the current detector and creates a new one.

Returns: MediaPipeSolution

Field	Type	Description
frameProcessor	ReadonlyFrameProcessor	Pass to <Camera frameProcessor={...}>.
cameraViewLayoutChangeHandler	(e: LayoutChangeEvent) => void	Pass to the camera’s onLayout.
cameraDeviceChangeHandler	(device?: CameraDevice) => void	Call when the active camera device changes.
cameraOrientationChangedHandler	(o: Orientation) => void	Pass to onOutputOrientationChanged.
resizeModeChangeHandler	(mode: 'cover' | 'contain') => void	Call with the camera’s resize mode.
cameraViewDimensions	{ width: number; height: number }	Current size of the camera view.

MediapipeCamera wires all of these up for you.

Example

const solution = usePoseDetection(
  { onResults, onError },
  RunningMode.LIVE_STREAM,
  'pose_landmarker_lite.task',
  { numPoses: 1, delegate: Delegate.GPU, fpsMode: 15 }
);