# Third-party assets

MediaPipe Tasks Vision 0.10.14: Copyright The MediaPipe Authors.
The npm package declares Apache-2.0 in the included tasks-vision-package.json.
The full upstream license from release v0.10.14 is mediapipe-LICENSE.txt.
Runtime JavaScript and WebAssembly files are unmodified copies from that package.

The Face Landmarker model bundle float16/1 is downloaded from Google's public
MediaPipe model storage. The official Face Landmarker guide identifies three
components: BlazeFace short-range, FaceMesh V2, and Blendshape V2. The official
model cards supplied here each state Apache License, Version 2.0 on their first
page. The full license text is supplied as Apache-2.0.txt. The bundle is unmodified.

Evidence supplied locally:

- face-detector-model-card.pdf (Google; June 9, 2021).
- face-mesh-v2-model-card.pdf (Google; September 15, 2022).
- blendshape-v2-model-card.pdf (Google; November 11, 2022).
- tasks-vision-package.json (upstream npm package metadata).

Official guide: https://developers.google.com/edge/mediapipe/solutions/vision/face_landmarker
Code license: https://github.com/google-ai-edge/mediapipe/blob/v0.10.14/LICENSE
Model: https://storage.googleapis.com/mediapipe-models/face_landmarker/face_landmarker/float16/1/face_landmarker.task

Exact file hashes and source URLs are in assets-lock.json in the project root.
These records pin the bytes retrieved for this project; they are not an upstream
signed provenance attestation. Model-card dates are distinct from bundle version 1.

Selfie Multiclass Segmentation 256x256 float32/1 is an unmodified Google model.
Its official model card states Apache License, Version 2.0; see the local
multiclass-segmentation-model-card.pdf and Apache-2.0.txt. Confidence categories
used here are hair (1) and face-skin (3). The model is not an anatomical hairline
detector and does not provide pixel-perfect masks.
Guide: https://developers.google.com/edge/mediapipe/solutions/vision/image_segmenter
Model: https://storage.googleapis.com/mediapipe-models/image_segmenter/selfie_multiclass_256x256/float32/1/selfie_multiclass_256x256.tflite