FaceEmbedding class

Generates face embeddings (identity vectors) from aligned face crops.

Uses the MobileFaceNet model to produce 192-dimensional embedding vectors that represent a face's identity. These embeddings can be compared using cosine similarity to determine if two faces belong to the same person.

The model expects 112×112 RGB face images as input. For best results, faces should be aligned (eyes horizontal, centered, face fills the frame).

Usage

final embedding = await FaceEmbedding.create();

// Get embedding from an aligned face crop
final vector = await embedding(alignedFaceCrop);
print('Embedding dimension: ${vector.length}'); // 192

// Compare two embeddings
final similarity = FaceEmbedding.cosineSimilarity(vector1, vector2);
if (similarity > 0.6) {
  print('Same person!');
}

// Clean up
embedding.dispose();

Integration with FaceDetector

For the full face recognition pipeline, use FaceDetector.getFaceEmbedding which handles face detection, alignment, and embedding extraction:

final detector = FaceDetector();
await detector.initialize();

final faces = await detector.detectFacesFromBytes(imageBytes);
final embedding = await detector.getFaceEmbedding(faces.first, imageBytes);