Face Verification — Siamese Network
Upload two photos of faces. A Siamese CNN embeds each face into a 32-dimensional vector; the distance between the two embeddings decides whether they're the same person.
Trained from scratch on the Olivetti Faces dataset (40 people, 10 images each). For best results, use clear, front-facing, well-lit photos.
Model card: Siamese CNN + contrastive loss, trained on Olivetti Faces (AT&T Laboratories Cambridge). Held-out pair accuracy: 100%.