Owner: Stephen. In mixed reality, the headset needs to know where the volunteer’s torso is. services/registration takes a Quest camera JPEG and returns body landmarks with Google’s MediaPipe Pose Landmarker Lite (float16, v1), on the CPU. Unity owns everything else: camera opt-in, calibration, projecting onto the table, fitting the torso, confidence checks and hiding anatomy when the fit isn’t valid. Shoulders and hips drive the fit, which is accepted after three stable observations.

​
Run it

The model download is explicit and checked against a pinned SHA-256. The server only binds to loopback; the headset reaches it over USB with adb reverse tcp:8790 tcp:8790.

​
Privacy

It stores no frames, landmarks, participant identifiers or request logs. Images exist only in memory while a request is processed. If no person or more than one person is detected, the result is empty and invalid.