Realtime computer-vision + graphics demos built to wow.
- Tilt your hand to steer the globe.
- Pinch (thumb + index) to expand it with a glow burst.
- Open palm (4–5 fingers) to trigger burst rings.
Tech:
- Hand tracking: MediaPipe Tasks Vision
HandLandmarker - Rendering: Three.js
- App: Vite + React + TypeScript
From vision-projects/:
npm install
npm run devThen open the URL printed in the terminal and allow camera access.
See USER_INSTRUCTIONS.md for the full guide.
Quick start:
-
Click Start camera and allow webcam permissions.
-
Try:
- Tilt: rotate your hand like a steering wheel → globe rotates/steers.
- Pinch: thumb + index pinch open/close → globe expands/shrinks.
- Open palm: show 4–5 fingers → burst rings animation.
- Two hands: show both hands; move them apart to expand and together to shrink.
The globe includes a small set of pinned labels (countries + major cities). This is a demo dataset and can be expanded to hundreds/thousands (with clustering) if you want.
- Works best in Chrome / Edge.
- If the model download is blocked by network policy, we can vendor the
.taskmodel file locally and load it frompublic/.