Vitomir Štruc

SCOUT reveals that face recognition templates can be semantically edited directly, unlocking new avenues for identity-aware manipulation without sacrificing accuracy.

Erasing someone's face from an AI is now possible by nudging its identity embedding towards a similar-looking "anchor" face, rather than retraining the entire model.

Open-source MLLMs can spot face morphing attacks with surprising accuracy, beating specialized systems without any training.