About Doppelmap
Doppelmap maps face-similarity between public figures. Every photograph comes from Wikimedia Commons and is compared with a face-recognition model; each person is reduced to a single consensus face, and UMAP lays those out so people who look alike sit near each other.
Pick anyone to see their closest matches. The ⇄ button shows the two photographs that actually match — not two arbitrary portraits, but the specific pair the model rates highest.
Similarity is raw cosine distance between faces. It runs lower than intuition suggests: photos of the same person typically score above 0.35, so two different people at 0.30 is already a striking resemblance.
Images are CC-licensed or public domain, credited on each photo. People are drawn from the most-viewed English Wikipedia biographies; fictional characters are excluded.
No personal data or photos are stored. The map holds only what is needed to draw it: names, the face crops shown here, and how alike the faces are. Voting stores only an anonymous id and which way the vote went, never a name, an email or a picture.