Live Recognition
Point a camera (or load a video). Faces are detected, embedded, and matched against the people you enroll — labeled live, in real time. Everything runs on your machine; no frame ever leaves the browser.
idle — models load on Start.
Archive Search
Build an offline face database — each enrolled face gets an identity dossier (name, DOB, ID#, status…). Then Identify: drop a probe photo and the system runs it against every record and pulls the matching profile, the classic "run it through the database" move. Or auto-cluster "these are the same person." All local to your browser (IndexedDB).
idle — models load on first use.
2D → 3D Reconstruction
Drop a single photo of a face. 478 3D landmarks are extracted and lifted into a rotatable wireframe mesh — a flat photo becomes a 3D face you can spin. Drag to rotate, scroll to zoom.
◉ SOURCE
idle — drop a face photo, or use ◈ Capture → 3D on the Live tab.
Built as a proof-of-concept / mockup. Recognition identifies people — use it on your own face or with consent. Nothing is uploaded; all processing is local to this browser.