Facial embedding

📖 Full definition

A facial embedding is a 128- to 512-dimensional vector that summarises the geometric "signature" of a face. Two similar faces produce embeddings close in vector space; two different faces produce distant embeddings. It is the core of all modern face recognition AIs (FaceNet, ArcFace, DeepFace).

💡 Concrete example

When you upload your photos to Look Like Me, our AI computes an embedding for you, your father and your mother. The resemblance score is then the cosine distance between your embedding and each parent's.

🔄 Synonyms

face vector · feature vector · face encoding

🔗 See also

• ArcFace →
• Similarity score →
• Deep learning →