# Face Embedding Models (TFLite)

This app expects a **face embedding** model in TFLite format, used for on-device
face recognition (cosine similarity between embeddings).

Supported asset paths (the app tries them in order):
- `assets/models/facenet.tflite` (commonly 160x160 input)
- `assets/models/mobilefacenet.tflite` (commonly 112x112 input)
- `assets/models/mobile_face_net.tflite` (commonly 112x112 input)

`facenet.tflite` has been replaced with a real FaceNet 128D embedding model
from `shubham0204/OnDevice-Face-Recognition-Android` (`app/src/main/assets/facenet.tflite`).
The source project documents the model as FaceNet 160x160 input with 128-dimensional
embeddings and is published under Apache-2.0.

`mobilefacenet.tflite` and `mobile_face_net.tflite` may still be placeholders;
the app tries `facenet.tflite` first.