# RadVeil

Copyright 2026 Piotr Gródek and the RadVeil contributors

Licensed under the Apache License, Version 2.0.

This product includes the MediaPipe BlazeFace (short range) and Face Mesh v2
model weights, Copyright Google LLC, licensed under the Apache License,
Version 2.0 (https://github.com/google-ai-edge/mediapipe). The files were
modified by RadVeil: exported from TFLite to RadVeil's own network format
(.rvnn) with unchanged float16 weights (crates/radveil-core/models/, see
README.md there). They power the built-in face detector. "MediaPipe" is used only to identify the models; RadVeil is not a MediaPipe or Google product.

This product includes the SynthStrip model weights (A. Hoopes, J. S. Mora,
A. V. Dalca, B. Fischl, M. Hoffmann; https://synthstrip.io), used under the
MIT License as offered by the SynthStrip authors. The file was modified by
RadVeil: converted from the PyTorch checkpoint with weights rounded to
float16 (crates/radveil-core/models/synthstrip.1.rvss; license and
provenance in models/LICENSE-synthstrip). It powers brain extraction
(`deskull`, method "synthstrip"). If you publish results, please cite
Hoopes et al., NeuroImage 260 (2022) 119474.

This product includes third-party Rust crates under permissive licenses
(MIT, Apache-2.0, BSD, ISC, Zlib and similar). Their notices and license
texts are in THIRD_PARTY_LICENSES.txt.

This product optionally uses dlib (Boost Software License) for face
detection at runtime, via the separately installed `dlib` Python package.
No dlib model file is distributed with this product. The default dlib
landmark model, downloaded by the user with radveil.fetch_predictor(), is
dlib's 5-point predictor, published under CC0-1.0. dlib's 68-point models
(iBUG 300-W training data, commercial use excluded) are used only when the
user supplies one explicitly. See docs/legal.md.
