Similarity score

📖 Full definition

The similarity score (or cosine similarity score) is computed from the cosine distance between two facial embeddings. It ranges from -1 (totally opposite) to 1 (identical). For consumer use, this value is converted to a percentage: 0% = very different faces, 100% = identical faces.

💡 Concrete example

If Look Like Me gives you a score of 78% dad and 65% mom, it means the cosine distance between your embedding and your father's is smaller (closer) than with your mother's.

🔄 Synonyms

cosine similarity · resemblance score

🔗 See also