About Hologram Workshop

A browser tool that turns a phone scan into a hologram you can handle with your bare hands. This page is an honest account of what it does, what it doesn't do yet, and how it works.

What it is

You scan a real object or room with a phone's LiDAR camera. The page shows it as a glowing hologram, and your webcam watches your hands so you can grab it, turn it, resize it, take it apart and measure it, the way you'd handle the real thing. A mouse does everything too.

It is a student project, free and open source: no install, no account, no server doing the work.

What it does

  • Hand control: aim with a pointing finger, click by pinching your other hand, fist to grab, two hands to tilt, scale or explode, clap to reset.
  • True-size measuring: the scan keeps its real-world scale, so the tape measure and size readout are in real centimetres.
  • Your own scans: the Platform opens GLB, glTF, OBJ and PLY files and splits them into objects and parts automatically.
  • A library: scenes and their versions are saved in your browser and shown as a ring you can spin.
  • Polygon lens and pins: look at the mesh under a lens, hide or mark faces without changing the scan, and pin pieces so they can't be moved by accident.
  • Practice: a short calibration and drills that arm one gesture at a time, so you can learn each move on its own.

What it does not do (yet)

  • Webcam tracking isn't perfect. Poor light, a hand at the edge of the frame or a hand turned side-on can make the cursor wobble or drop. Good front light helps most.
  • It has mostly been tested by one person. Gesture thresholds are tuned on one set of hands, one laptop camera and one room.
  • Some gestures are still coming. The ✌ tool wheel and the 👎 undo gesture are designed but not built; undo works with U or Ctrl+Z today.
  • Scan completion and photo → 3D run offline. They are Python tools that run on a computer, not in the browser. In the browser a photo shows as a flat preview.
  • Not built for phones. The pages open on a phone, but hand control needs a laptop or desktop with a webcam.

How it works

  1. Hand tracking: Google's MediaPipe finds 21 points on each hand in every webcam frame, inside your browser.
  2. Smoothing: a One Euro filter steadies the points without adding lag when you move fast.
  3. Gestures: each pose has one meaning. A short pause between different gestures stops one from bleeding into the next.
  4. The hologram: three.js draws the scan with a holographic material. It is tested so nothing flashes more than three times a second.

Scan completion

A phone scan only sees what the camera saw: a chair's underside, the back of a sofa against a wall, and gaps the scanner skipped come out missing. An offline Python tool fills those in from the measured surfaces around them (thickness, symmetry and flat planes).

The real scan always wins. Measured faces are never moved, and every filled-in surface is marked as inferred, drawn with an amber hatch, so you always know which parts were measured and which were guessed. On the sample chair the filled underside matches the real one to within 2 cm on about 98% of its area.

Privacy

The camera feed is processed in your browser and never leaves your device. No video, image, hand data or scan is uploaded. There are no accounts, ads or analytics. Your saved scenes live in this browser's storage; clearing site data removes them.

Works in

Recent Chrome, Edge and Safari on a laptop or desktop with a webcam. The camera needs a secure page (this site is served over HTTPS) and your permission once.

Credits and licences