focusedgaze: third-party components and model licensing
========================================================

The focusedgaze SOURCE CODE is licensed under the MIT License (see LICENSE).
That licence covers this project's own code and nothing else.


0. Author
---------

  Muhammad Asif Khan   author, maintainer
    https://github.com/muhammad-asifkhan

This package does NOT distribute, bundle, mirror, or automatically download any
gaze model weights. Read this file before using it for anything commercial.


1. Gaze model weights: RESTRICTED
----------------------------------

focusedgaze runs an ONNX export of L2CS-Net trained on the Gaze360 dataset
(commonly distributed as `L2CSNet_gaze360.pkl`, converted locally to
`l2cs_gaze360.onnx`).

  * L2CS-Net source code: MIT License
    https://github.com/Ahmednull/L2CS-Net

  * Gaze360 dataset and derived models: NON-COMMERCIAL RESEARCH USE ONLY
    https://github.com/erkil1452/gaze360
    The Gaze360 authors state: "The usage of the dataset and the code is for
    non-commercial research use only."

Model weights trained on Gaze360 are normally treated as a derived work and
carry the same restriction. Therefore:

  * This project does not distribute those weights.
  * This project does not host or mirror them.
  * `focusedgaze download-models` deliberately does NOT fetch them. It prints
    the upstream source and this restriction, and stops.
  * You obtain the weights yourself, from the official L2CS-Net distribution
    linked in that project's README, and convert them locally with
    `focusedgaze export-onnx`.
  * You are responsible for your own compliance with the Gaze360 terms.

The MIT classifier on this package describes THIS PACKAGE'S CODE. It makes no
claim about the licence of any model you choose to load into it.

This is a conservative reading of the upstream terms, not legal advice. If a
qualified review finds the restriction does not apply, this policy can be
relaxed; it will not be tightened.


2. MediaPipe face landmarker: permissive
-----------------------------------------

`face_landmarker.task` is published by Google under the Apache License 2.0 and
is downloaded automatically from Google's own hosting, with SHA-256
verification. https://ai.google.dev/edge/mediapipe


3. Runtime dependencies
-----------------------

  MediaPipe        Apache-2.0
  OpenCV (opencv-python)  Apache-2.0
  NumPy            BSD-3-Clause
  ONNX Runtime     MIT
  platformdirs     MIT
  scikit-learn     BSD-3-Clause   (calibration extra; fitting only)
  SciPy            BSD-3-Clause   (calibration extra; optional)
  websockets       BSD-3-Clause   (server extra)

Each remains under its own licence. Installing focusedgaze does not relicense
any of them.
