# MediaPipe third-party notices

PicSelec distributes the following unmodified Google MediaPipe model assets
for local, on-device inference:

- `blaze_face_full_range.tflite`
  - Source: https://storage.googleapis.com/mediapipe-models/face_detector/blaze_face_full_range/float16/1/blaze_face_full_range.tflite
  - SHA-256: `3698B18F063835BC609069EF052228FBE86D9C9A6DC8DCB7C7C2D69AED2B181B`
  - Model: MediaPipe BlazeFace Full Range
- `blaze_face_full_range_sparse.tflite`
  - Source: https://storage.googleapis.com/mediapipe-assets/face_detection_full_range_sparse.tflite
  - SHA-256: `2C3728E6DA56F21E21A320433396FB06D40D9088F2247C05E5635A688D45DFE1`
  - Model: MediaPipe BlazeFace Sparse Full Range
- `face_landmarker.task`
  - Model: MediaPipe Face Landmarker

BlazeFace authors listed by the official model cards include Valentin
Bazarevsky, Yury Kartynnik, and Artsiom Ablavatski at Google.

The BlazeFace model cards license these models under the Apache License,
Version 2.0:

- Full-range sparse model card: https://storage.googleapis.com/mediapipe-assets/MediaPipe%20BlazeFace%20Sparse%20Model%20Card%20%28Full%20Range%29.pdf
- Short-range model card and common license statement: https://storage.googleapis.com/mediapipe-assets/MediaPipe%20BlazeFace%20Model%20Card%20%28Short%20Range%29.pdf
- Apache License 2.0: https://www.apache.org/licenses/LICENSE-2.0

MediaPipe Tasks Vision code and WebAssembly runtime are distributed under
the Apache License, Version 2.0. The runtime source and license are available
at https://github.com/google-ai-edge/mediapipe.

- `pose_landmarker_lite.task`
  - Model: MediaPipe Pose Landmarker Lite, float16 version1; downloaded2026-10-12
  - Source: https://storage.googleapis.com/mediapipe-models/pose_landmarker/pose_landmarker_lite/float16/1/pose_landmarker_lite.task
  - SHA-256: `59929E1D1EE95287735DDD833B19CF4AC46D29BC7AFDDBBF6753C459690D574A`
  - Bytes: 5,777,746
  - Official guide: https://developers.google.com/edge/mediapipe/solutions/vision/pose_landmarker
  - Model card: https://storage.googleapis.com/mediapipe-assets/Model%20Card%20BlazePose%20GHUM%203D.pdf
  - License: Apache License, Version2.0 (model card, page2)
  - Local pose/visible-silhouette candidates only; no identity recognition or anatomical measurement claim.

- `body-vision-1.1.0/wasm/vision_wasm_module_internal.{js,wasm}`
  - Unmodified runtime assets from npm `@mediapipe/tasks-vision@1.1.0`, installed through the isolated `@mediapipe/body-vision` npm alias.
  - Source/license: https://github.com/google-ai-edge/mediapipe (Apache License2.0)
  - Used only by the optional body analysis module Worker. Existing face model/runtime1.0.0 remains paired with its original assets.
  - ES module loader selected explicitly. Non-SIMD module initialization is unavailable with this vendor distribution and returns analysis failure; no universal old-browser support claim.

## Hand Landmarker (optional natural face shape protection)

- Model: Google MediaPipe Hand Landmarker float16, version 1.
- Source: https://storage.googleapis.com/mediapipe-models/hand_landmarker/hand_landmarker/float16/1/hand_landmarker.task
- Guide: https://developers.google.com/edge/mediapipe/solutions/vision/hand_landmarker/web_js
- Model card/license: https://storage.googleapis.com/mediapipe-assets/Model%20Card%20Hand%20Tracking%20(Lite_Full)%20with%20Fairness%20Oct%202021.pdf (Apache License 2.0, page 2).
- File: hand_landmarker.task; 7,819,105 bytes; SHA-256 fbc2a30080c3c557093b5ddfc334698132eb341044ccee322ccf8bcf3607cde1.
- Runtime: installed @mediapipe/body-vision alias to @mediapipe/tasks-vision 1.1.0, same-origin matching WASM. Existing face 1.0 runtime remains separate.
- Landmarks and convex protection covers are estimates, not segmentation. Occluded/gloved hands and jewelry may be missed; manual protection remains available. No photo upload or off-origin telemetry is allowed.

## Selfie Multiclass Segmentation (optional neck/arm skin tone)

- File: `selfie_multiclass_256x256.tflite`; 16,371,837 bytes.
- SHA-256: `c6748b1253a99067ef71f7e26ca71096cd449baefa8f101900ea23016507e0e0`.
- Pinned source: https://storage.googleapis.com/mediapipe-models/image_segmenter/selfie_multiclass_256x256/float32/1/selfie_multiclass_256x256.tflite
- Model card/license: https://storage.googleapis.com/mediapipe-assets/Model%20Card%20Multiclass%20Segmentation.pdf (Apache License 2.0).
- Official guide: https://developers.google.com/edge/mediapipe/solutions/vision/image_segmenter
- CPU in an isolated worker with the existing `body-vision-1.1.0` runtime; all model/WASM requests remain same-origin. No photo upload or off-origin telemetry.
- Six classes: background, hair, body skin, face skin, clothes, other/accessories. These are estimates, not pixel-perfect masks. High-confidence skin is intersected with pose ownership and neck/arm regions; uncertain edges abstain and protection brushes remain available.
