Reproduce “Specify Privacy Yourself: Assessing Inference-Time Personalized Privacy Preservation Ability of Large Vision-Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xingqi Wang, Xiaoyuan Yi, Xing Xie, Jia Jia
The authors introduce Inference-Time Personalized Privacy Protection and evaluate it with SPY-Bench, a benchmark of 21 vision-language models. They fine-tune Qwen2 VL 7B Instruct using paired private and non-private instructions. With distractors and scenarios, NCA-P reaches 96.88 ICS and 88.11 HMS, compared with 2.75 and 12.33 for the untuned model. These results show performance on this benchmark, not dependable protection beyond it.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.