Face & Emotion Tracker

PROJECT_ABSTRACT

Face & Emotion Tracker, has no offline pipeline, every frame is captured from getUserMedia and run through face-api.js live, on the visitor's own machine, the moment the page loads. TinyFaceDetector locates faces, a 68-point landmark model traces jaw/eyebrows/eyes/nose/mouth (iBUG 300-W layout), and a 7-class expression classifier scores neutral/happy/sad/angry/fearful/disgusted/surprised. Runs on TensorFlow.js with WebGL acceleration (CPU fallback), model weights pinned in public/models/ so there's zero external CDN dependency at runtime. Ships as a pure static bundle.

KEY_CAPABILITIES

• ▶Fully client-side inference
• ▶no offline precompute step
• ▶every frame processed live via getUserMedia + requestAnimationFrame.
• ▶68-point face mesh (iBUG 300-W) plus dominant-emotion label drawn per frame onto a canvas sized to native video resolution.
• ▶Pinned model weights in public/models/ remove any external CDN dependency at runtime; WebGL backend with automatic CPU fallback.
• ▶Rolling 20-frame FPS counter and live backend indicator. graceful inline message when camera access or getUserMedia is unavailable.