facenet-128d-embeddings-curse-mitigation
IN premise — summaries/2026/08/24/wiki-Curse_of_dimensionality-chunk-3.md
Created 2026-08-24T17:11:09+00:00
FaceNet (Schroff et al., 2015) uses deliberately low-dimensional 128-D face embeddings, reflecting the practical trade-off between representational richness and the curse of dimensionality in learned embedding design.
Summary
When FaceNet turns a face into a 128-number fingerprint, it is making a deliberate choice to keep that number low rather than pushing to thousands of dimensions. This matters because it acknowledges a real limit: adding more dimensions to an embedding does not keep making it better, since high-dimensional spaces make distances unreliable and demand far more data to fill in, so a modest dimensionality is often the sweet spot between capturing enough identity detail and keeping the geometry usable.