Steering Vision-Language Models for Visual-Contextual Privacy in Location Disclosure
An inference-time route-and-steer framework is proposed: an ordinal granularity router maps visual-contextual privacy representations to a disclosure route, and a disclosure-level-conditioned steering module applies route-specific hidden-state interventions during decoding.
Wen-Yue Chen, Yu-Han Cheng, Ruixin Yang et al.
· 0 citations