KARARTISOLUTIONS beyzakararti.com Sabyeza

Gesture 3D Viewer:
turn the model by hand

An open source desktop application that overlays local 3D models and presentation images over the webcam image, controlling them all with hand gestures. Hold a mechanical part in the air and turn it, zoom in with both hands, secure it with the presentation lock, and record while you explain.

GLB · STL · OBJnative 3D model support
MediaPipehand, face, pose and person segmentation
127.0.0.1everything is local, the camera does not go out
open sourcePublicly available on GitHub

Gestures

There is no mouse,
there is a hand

You hold the model with your thumb and index finger, open and turn your palm, and zoom in by opening and closing both hands. Movements are softened against tremors; Dead zone and limitation prevent sudden jumps.

Grab and carry

Pinch

When the thumb-index finger tip approaches, a beam is shot at the model from the middle point of the fingers; If the model hits, it is grabbed and dragged in the camera plane. Separate hold and release thresholds prevent boundary jitter.

Rotate and zoom

Open palm · two hands

The movement of the palm center in one hand is converted into rotation, and the logarithmic distance change between the palm centers in both hands is converted into zoom.

presentation lock

Spock/Vulcan

The Vulcan salute, held for approximately 900 ms, locks the model in presentation position; To trigger again, the hand is lowered briefly. When the person moves in front of the model, the segmentation mask naturally covers the model.

What is it used for?

  • Examining the GLB, STL and OBJ mechanical parts on the camera image (e.g. FRAGWATT's STL files turning with hand gesture during presentation).
  • Rotating models and pointing out their parts by hand in technical training.
  • Fixing article pages or diagrams on both shoulders.
  • Recording WebM videos with narration for lectures, science communication, and online meetings.
  • Developing gesture-based human–computer interaction experiments.

Architecture and security

  • Electron + Three.js interface; Python FastAPI WebSocket and MediaPipe Holistic tracking only at 127.0.0.1:8765.
  • Frames 480 × 270 JPEG, up to 8 FPS, single frame at a time; raw frames are not written to disk.
  • contextIsolation, sandbox, nodeIntegration: false; preload only opens fixed IPC channels; CSP blocks remote code.
  • The camera, microphone and recording are turned on only after the user presses the button.
Status: experimental but functional; It is developed as open source. The roadmap includes gesture calibration, depth camera support, STEP/IGES and CAD assembly tree, cross-section and measurement tools, virtual camera for meeting platforms, and AR glasses support. Single camera distance calculation is approximate.