Spaces:
Sleeping
Sleeping
File size: 5,033 Bytes
cf93910 c476eae cf93910 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 | import { useEffect, useRef, useState, useCallback } from 'react';
import {
HandLandmarker,
FilesetResolver,
type HandLandmarkerResult,
type NormalizedLandmark,
} from '@mediapipe/tasks-vision';
import { normaliseLandmarks } from '../lib/landmarkUtils';
export interface MediaPipeState {
landmarks: number[] | null; // 63-float normalised vector
rawLandmarks: NormalizedLandmark[] | null; // 21-point raw result (for canvas drawing)
handedness: 'Left' | 'Right' | null;
isDetecting: boolean;
isLoading: boolean;
error: string | null;
startDetection: (video: HTMLVideoElement) => void;
stopDetection: () => void;
}
const WASM_URL =
'https://cdn.jsdelivr.net/npm/@mediapipe/tasks-vision@latest/wasm';
const MODEL_URL =
'https://storage.googleapis.com/mediapipe-models/hand_landmarker/hand_landmarker/float16/1/hand_landmarker.task';
/**
* Hook that drives MediaPipe HandLandmarker inference on a video element.
* Runs at ~30 fps using requestAnimationFrame.
*/
export function useMediaPipe(): MediaPipeState {
const landmarkerRef = useRef<HandLandmarker | null>(null);
const rafRef = useRef<number>(0);
const lastTsRef = useRef<number>(0);
const [landmarks, setLandmarks] = useState<number[] | null>(null);
const [rawLandmarks, setRawLandmarks] = useState<NormalizedLandmark[] | null>(null);
const [handedness, setHandedness] = useState<'Left' | 'Right' | null>(null);
const [isDetecting, setIsDetecting] = useState(false);
const [isLoading, setIsLoading] = useState(false);
const [error, setError] = useState<string | null>(null);
// initialise on mount
useEffect(() => {
let cancelled = false;
setIsLoading(true);
(async () => {
try {
const vision = await FilesetResolver.forVisionTasks(WASM_URL);
// Try GPU first for best performance; fall back to CPU if unavailable.
let hl: HandLandmarker | null = null;
try {
hl = await HandLandmarker.createFromOptions(vision, {
baseOptions: { modelAssetPath: MODEL_URL, delegate: 'GPU' },
runningMode: 'VIDEO',
numHands: 1,
minHandDetectionConfidence: 0.4,
minHandPresenceConfidence: 0.4,
minTrackingConfidence: 0.4,
});
} catch {
console.warn('GPU delegate unavailable, falling back to CPU.');
hl = await HandLandmarker.createFromOptions(vision, {
baseOptions: { modelAssetPath: MODEL_URL, delegate: 'CPU' },
runningMode: 'VIDEO',
numHands: 1,
minHandDetectionConfidence: 0.4,
minHandPresenceConfidence: 0.4,
minTrackingConfidence: 0.4,
});
}
if (!cancelled) {
landmarkerRef.current = hl;
setIsLoading(false);
}
} catch (err) {
if (!cancelled) {
console.error('MediaPipe init error', err);
setError('Failed to load hand detection model. Check network.');
setIsLoading(false);
}
}
})();
return () => {
cancelled = true;
cancelAnimationFrame(rafRef.current);
landmarkerRef.current?.close();
};
}, []);
const startDetection = useCallback((video: HTMLVideoElement) => {
if (!landmarkerRef.current) return;
setIsDetecting(true);
const detect = (now: number) => {
if (!landmarkerRef.current || !video || video.paused || video.ended) {
rafRef.current = requestAnimationFrame(detect);
return;
}
// Throttle to ~30 fps
if (now - lastTsRef.current < 33) {
rafRef.current = requestAnimationFrame(detect);
return;
}
lastTsRef.current = now;
let result: HandLandmarkerResult;
try {
result = landmarkerRef.current.detectForVideo(video, now);
} catch {
rafRef.current = requestAnimationFrame(detect);
return;
}
if (result.handednesses.length > 0 && result.landmarks.length > 0) {
const raw = result.landmarks[0]; // NormalizedLandmark[]
const hand = result.handednesses[0][0].categoryName as 'Left' | 'Right';
try {
const flat = normaliseLandmarks(raw);
setLandmarks(flat);
setRawLandmarks(raw);
setHandedness(hand);
} catch {
setLandmarks(null);
setRawLandmarks(null);
}
} else {
setLandmarks(null);
setRawLandmarks(null);
setHandedness(null);
}
rafRef.current = requestAnimationFrame(detect);
};
rafRef.current = requestAnimationFrame(detect);
}, []);
const stopDetection = useCallback(() => {
cancelAnimationFrame(rafRef.current);
setIsDetecting(false);
setLandmarks(null);
setRawLandmarks(null);
setHandedness(null);
}, []);
return {
landmarks,
rawLandmarks,
handedness,
isDetecting,
isLoading,
error,
startDetection,
stopDetection,
};
}
|