This project reimplements the liveness detection model from Glance
(https://github.com/jonnyoo/glance) by Jonathan Zhou, MIT licensed.

The five-cue deny/confirm design, the tuning constants, and the reasoning
recorded in the comments of src/glanced/liveness/ originate there. The Swift
implementation was not copied; each module was rewritten against MediaPipe,
OpenCV and NumPy, and the deviations are documented at each site.

The ArcFace recognition model is from InsightFace
(https://github.com/deepinsight/insightface).

The lock screen's success animation (patches/omarchy-lock-faceid/
unlock-spin.png) is frames 16-45 of the unlock video shipped with Glance
(Resources/unlockanimation.mp4), cut to the inside of its ring and reduced
to an alpha mask so it can take the current theme's colour.

The onnxruntime landmarker (src/glanced/landmarker_onnx.py) runs Google's
MediaPipe BlazeFace short-range and Face Mesh V2 networks, taken unchanged
from face_landmarker.task (float16/1) and converted from TFLite to ONNX
(see packaging/MODELS.md). src/glanced/_data/face_geometry/ and
packaging/face_geometry/ hold MediaPipe's geometry pipeline metadata, copied
byte-for-byte. landmarker_onnx.py re-implements the behaviour of MediaPipe's
FaceLandmarker graph (ROI rules, One-Euro landmark smoothing, geometry
pipeline). Copyright 2023 The MediaPipe Authors, licensed under the Apache
License, Version 2.0: http://www.apache.org/licenses/LICENSE-2.0
