The Phantom Menace: Unmasking Privacy Leakages in Vision-Language Models
Simone Caldarella, Massimiliano Mancini, Elisa Ricci, and Rahaf Aljundi
🗞️News Accepted at Trust What You Learn Workshop @ ECCV 2024
TL;DR
- VLMs leak identity information, even when the vision-language alignment and the fine-tuning use anonymized data
- Context has little influence on identity leakage
- Simple, widely used anonymization techniques, like blurring, are not sufficient to address the problem
Read the full paper on Arxiv 