AI RESEARCH
NearID: Identity Representation Learning via Near-identity Distractors
arXiv CS.CV
•
ArXi:2604.01973v1 Announce Type: new When evaluating identity-focused tasks such as personalized generation and image editing, existing vision encoders entangle object identity with background context, leading to unreliable representations and metrics. We