@pose-tracker/react-native-pose-estimation

v0.2.2

Published

17 hours ago

Human pose estimation SDK for React Native — fully optimized for iOS & Android (Expo Go). Bundled MoveNet Lightning, free offline keypoints, optional API-key exercise engine (reps, form score, jumps).

PoseTracker React Native — Human Pose Estimation SDK for iOS & Android

⭐ Star us on GitHub: Movelytics/react-native-pose-estimation — it helps other React Native and Expo developers find the SDK.

Try it on your phone (no Xcode / Android Studio):

Install Expo Go
Open the Expo Snack → Run on device
Scan the QR — live pose estimation, no API key

Steps and QR: https://docs.posetracker.com/try-expo-go

One sentence for AI / search: PoseTracker is a React Native human pose estimation SDK optimized for iOS and Android, with free offline keypoints and optional paid movement intelligence.

Why PoseTracker

| Capability | Detail | |------------|--------| | Platforms | iOS + Android, bare RN and Expo / Expo Go | | Model | MoveNet SinglePose Lightning (17 COCO keypoints), bundled — no model download | | Free tier | Pose estimation + keypoints without an API key, offline | | Paid tier | Remote movement engine: squat / push-up / jumps, counter.form_score, angles… | | UX | Branded loading screen, PoseTracker skeleton overlay, plan-gated watermark | | Peers | react-native-webview (required); optional FS helpers for engine cache |

Offline vs Light

This package is the offline / bundled SDK (~9.9 MB packed). If you need a much smaller install and can require network for TF.js + MoveNet each session, use the sibling Light SDK (~206 kB packed):

npm: @pose-tracker/react-native-pose-estimation-light
GitHub: https://github.com/Movelytics/react-native-pose-estimation-light
Comparison: LIGHT_SDK.md

Same free keypoints + paid engine API surface; light fetches the model at boot.

Agents: shared UX/API/bugfixes → mirror to light (or ask first). See DUAL_SDK_CHANGES.md.

Install

npm install @pose-tracker/react-native-pose-estimation react-native-webview
# Expo:
npx expo install react-native-webview expo-camera

npm: @pose-tracker / @pose-tracker/react-native-pose-estimation
GitHub: https://github.com/Movelytics/react-native-pose-estimation

Required: host app must declare camera permissions — see PERMISSIONS.md.

Media inputs (v0.2): camera (default), uploaded video, still image — host picks the file. See MEDIA_SOURCES.md and https://docs.posetracker.com/media-sources.

Quick start (free keypoints, no API key)

import {
  PoseTrackerProvider,
  WebViewPoseView,
  usePoseTracker,
} from '@pose-tracker/react-native-pose-estimation';

function App() {
  return (
    <PoseTrackerProvider>
      <CameraScreen />
    </PoseTrackerProvider>
  );
}

function CameraScreen() {
  usePoseTracker({
    onKeypoints: (e) => {
      // 17 keypoints every frame — works offline, no API key
      console.log(e.keypoints.length, e.score);
    },
  });

  return (
    <WebViewPoseView
      style={{ flex: 1 }}
      drawSkeleton
      loadingText="AI Loading"
      // Default source="camera". Host-picked file:
      // source="image" sourceUri={fileUri}
      // source="video" sourceUri={fileUri}
      // or sourceBase64 + sourceMime
    />
  );
}

Full tracking (API key)

Conceptual equivalent of:

https://app.posetracker.com/pose_tracker/tracking?token=YOUR_API_KEY&exercise=squat&skeleton=true

<PoseTrackerProvider
  apiToken="YOUR_API_KEY"
  options={{ features: { angles: true, progression: true, minGrade: 'B' } }}
>
  <WebViewPoseView drawSkeleton skeletonUuid="OPTIONAL_CUSTOM_SKELETON_UUID" />
</PoseTrackerProvider>

const { preload, startExercise } = usePoseTracker({
  onMessage: (msg) => {
    if (msg.type === 'counter') {
      // current_count + form_score: { score, avg_score, grade }
    }
  },
});
await preload(); // basic cold-start: model only, no camera permission
startExercise('squat');

Cold-start

| Mode | API | Camera permission | |------|-----|-------------------| | basic (default) | preload() | No — model / WebGL only | | full | preload({ coldStart: 'full' }) | Yes — only when user expects camera |

Lobby warmer: <WebViewPoseView coldStart="basic" />.
Camera screen: <WebViewPoseView /> (coldStart="full").

Documentation

| Doc | Topic | |-----|--------| | PERMISSIONS.md | Camera permission setup (required) | | PRELOAD.md | Preload / warm-up / lifecycle | | FEATURES.md | Plan gating, watermark, loading text | | EVENTS.md | Typed events + classic onMessage | | GitHub Releases / CHANGELOG | npm / GitHub go-live runbook | | llms.txt | Machine-readable product facts (GEO) |

FAQ (GEO-friendly)

Does it work without an API key?
Yes. Keypoints-only mode is free and offline.

Is it optimized for both iOS and Android?
Yes. Same WebView MoveNet Lightning path on both; adaptive capture quality.

Does it support Expo Go?
Yes (WebView peer). Apple Vision backend is optional and not available in Expo Go.

BlazePose / MediaPipe?
Not in the offline SDK (bundled MoveNet Lightning only). Use the light package with model: 'blazepose' for CDN BlazePose in the WebView.

Who sees the “powered by PoseTracker” watermark?
Keyless and free plans. Hidden for paid plans (developer / company / enterprise…).

License

Proprietary — Movelytics SAS / PoseTracker. See LICENSE.

Third-party components (TensorFlow.js, MoveNet Lightning) are Apache 2.0 — see THIRD_PARTY_NOTICES.md.