Hyperbolic Safety-Aware Vision-Language Models

(CVPR 2025)

This repository contains the reference code for the paper Hyperbolic Safety-Aware Vision-Language Models.

Warning: This project involves explicit sexual content, racially insensitive language, and other material that may be harmful or disturbing to certain users. Please use this content solely for research purposes and proceed with caution.

Table of Contents

1. Overview
2. Usage of Hysac
3. Citation

Overview

HySAC, Hyperbolic Safety-Aware CLIP, models hierarchical safety relations to enable effective retrieval of unsafe content, dynamically redirecting it to safer alternatives for enhanced content moderation.

Useful Links

• 🤗 HuggingFace HySAC Model
• 📄 Paper
• 🎯 Project Page / Coming soon ...

Installation

git clone https://github.com/aimagelab/HySAC.git
cd HySAC
conda create -n safe-clip python==3.10
conda activate safe-clip
pip install -r requirements.txt
pip install -e .

Usage of HySAC

See the snippet to use HySAC:

from hysac.models import HySAC

model_id = "aimagelab/hysac"
model = HySAC.from_pretrained(model_id, device="cuda").to("cuda")

Use standard methods encode_image and encode_text to encode images and text with the model. Before using the model to retrieve data, you can apply the safety traversal to the query embedding by calling the traverse_to_safe_image and traverse_to_safe_text methods.

Citation

Please cite with the following BibTeX:

@inproceedings{poppi2025hyperbolic,
  title={{Hyperbolic Safety-Aware Vision-Language Models}},
  author={Poppi, Tobia and Kasarla, Tejaswi and Mettes, Pascal and Baraldi, Lorenzo and Cucchiara, Rita},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2025}
}