Privacy-Leakage Mitigation in Vision Language Models

Overview

This is the official repository for our EACL Findings 2026 paper: Defeating Cerberus: Privacy-Leakage Mitigation in Vision Language Models.

Install

To install necessary packages, run:

pip install -r requirements.txt

Basic Usage and Examples

For hands-on examples on how to evaluate both LLM and multimodal models, please refer to our example notebooks. Additionally, results evaluation scripts can be found in dataset directory

Datasets

Multimodal datasets and dataset generation scripts can be found in dataset directory.

Note: all third-party content with their respective copyrights. These are used here as examples. You can replace them with your own content.

Citation

If you find this useful in your research, please consider citing:

TBD