MediaPipe Tasks Vision 1.2.0 — Google, Apache License 2.0 (LICENSE).
Source: https://github.com/google-ai-edge/mediapipe
Package: https://www.npmjs.com/package/@mediapipe/tasks-vision/v/1.2.0
Face Landmarker float16 model, revision 1:
https://storage.googleapis.com/mediapipe-models/face_landmarker/face_landmarker/float16/1/face_landmarker.task
Model documentation: https://developers.google.com/edge/mediapipe/solutions/vision/face_landmarker
Face landmarks do not directly estimate gaze. Tanym's per-user regression,
independent validation and visualization are separate application code.
All inference is local. No frame upload, recognition of identity or emotion analysis.

## EyeGestures V4
The adapted gaze engine is supplied separately in [eyegestures/](./eyegestures/README.md),
with its GPL-3.0 license, upstream provenance, corresponding Rust source and build instructions.
MediaPipe locates landmarks; EyeGestures maps them to screen gaze after personal calibration.