What Changes and What Doesn't
Aging affects different facial components at different rates. Soft tissue, skin, subcutaneous fat, and muscle, changes substantially: skin loses elasticity and develops texture, fat redistributes and reduces in some areas, and facial volume changes over decades. Bone structure, by contrast, is largely stable after early adulthood. The mandible, orbital bones, and nasal cartilage retain their fundamental configuration for most of adult life.
Face recognition systems built on deep embeddings primarily capture structural geometry rather than surface texture, making them more robust to aging than systems that relied on texture features. The embedding of the same person photographed at 25 and at 55 will be more similar than a texture-based comparison would suggest, the bone geometry anchor keeps the embedding in roughly the same region of face space.
Cross-Age Matching in Practice
Cross-age face verification benchmarks test systems on pairs with large temporal gaps. State-of-the-art systems achieve above 90% accuracy on these benchmarks when gaps are under 20 years, with accuracy declining for larger gaps, particularly when childhood photos are compared against adult photos, since facial proportions change substantially during growth.
For Ollie, cross-age robustness means that a photo taken a decade ago will typically return the same top celebrity matches as a recent photo, the stable geometry signal dominates the changing surface signal. Photos from adolescence or early adulthood may diverge, because the face structure may not yet have fully developed.