Face clustering

After faces are turned into embeddings, clustering looks at which embeddings sit close together and treats each tight group as one person. The result is a set of "people albums" the organiser can name.

Good clustering has to handle a person appearing in very different lighting, angles, and expressions across a day, while keeping two similar-looking guests apart.

In Pixeva this powers Smart Albums, and it lets a guest jump straight to their named album instead of running a selfie search.