Modelpublic

paper-2229/openclip-llava

sourceHugging Facellama2updated 11mo agoView on Hugging Face

0likes13downloads

Model Card

OpenCLIP-LLaVA

OpenLVLM-MIA: A Controlled Benchmark Revealing the Limits of Membership Inference Attacks on Large Vision-Language Models

—OpenLVLM-MIA offers a controlled benchmark to reassess membership inference attacks (MIA) on large vision-language models beyond dataset-induced biases.
—The benchmark consists of a 6,000-image dataset with controlled member/non-member distributions and ground-truth membership at three training stages.
—On this setup, state-of-the-art MIA approaches perform at chance level, clarifying the true difficulty of the problem and motivating more robust privacy defenses.