Benchmark suite for CLIP membership inference

CLIP-MIA-Bench is a benchmarking suite for evaluating membership inference attacks (MIA) against CLIP-style vision–language models. It provides standardized baselines, scripts, and reproduction code accompanying the paper:

Rethinking Membership Inference Attacks for CLIP
Lluis Gomez, AAAI 2026

The goal of this repository is to facilitate reproducible, fair, and extensible evaluation of MIA methods under different CLIP model variants, datasets, and attack assumptions.