FaceNet

A 2015 Google model that revolutionised facial recognition via embeddings.

📖 Full definition

FaceNet, introduced by Google in 2015, was the first model to learn a facial embedding directly optimised for similarity (rather than classification). It reached 99.6% accuracy on the LFW benchmark. Its principle — "triplet loss" — is still used today.

💡 Concrete example

Google Photos uses FaceNet to automatically group photos of the same person in your library.

🔗 See also

🪞 Test resemblance with our AI

Start my free test →