handtrack.js

Why we included this project

Handtrack.js gives front-end developers a quick route to realtime hand detection in the browser without training a model or standing up a server. It treats hand tracking as an object detection problem, and the pretrained convolutional network it ships runs through TensorFlow.js, so you can draw bounding boxes around hands in a webcam feed with only a few lines of code. The focus is on prototyping: interaction designers, game hobbyists, and small teams experimenting with gesture-controlled web UIs will find it far more approachable than building out a full ML pipeline. Recent versions also classify poses like open, closed, pinch, point, and zoom, which is handy for drag-and-drop metaphors, touch-free controls, and other playful inputs. One caveat: the newer pose set is still being refined, so for anything close to production you may want to pin the more stable 0.0.13 release.