Computer vision / 2026 / Live in the browser
Fusion Webcam Tracker
Face, hands and 33-point pose tracked at once, fusing MediaPipe Tasks with parallel TensorFlow.js detectors on a GPU-first backend.
01
The problem
Running several detectors per frame normally collapses the frame rate.
02
How I approached it
Backend selection walks WebGPU then WebGL before falling back to CPU, detectors share a single camera texture, and per-detector budgets stop any one model from starving the others.
03
The outcome
Three tracking modalities in the same frame loop without HUD clutter or dropped frames.
Related work
Hand-tracked cursed-technique visualiser: MediaPipe landmarks drive a Three.js particle engine, with an ML classifier picking the technique in real time.
Four image filters written twice, once as a Canvas2D loop and once as a WGSL compute shader, running on the same webcam frame side by side with honest timings.
ASL fingerspelling to text from a webcam. It reads hand geometry instead of pixels, so all 26 letters work with no training, and two seconds of calibration per letter fits it to your own hand.