Saved in:
Bibliographic Details
Main Authors: Lee, Wonjun, Lee, Doehyeon, Choi, Eugene, Yu, Sangyoon, Yousefpour, Ashkan, Park, Haon, Ham, Bumsub, Kim, Suhyun
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.04757
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912498725158912
author Lee, Wonjun
Lee, Doehyeon
Choi, Eugene
Yu, Sangyoon
Yousefpour, Ashkan
Park, Haon
Ham, Bumsub
Kim, Suhyun
author_facet Lee, Wonjun
Lee, Doehyeon
Choi, Eugene
Yu, Sangyoon
Yousefpour, Ashkan
Park, Haon
Ham, Bumsub
Kim, Suhyun
contents Current Vision Language Models (VLMs) remain vulnerable to malicious prompts that induce harmful outputs. Existing safety benchmarks for VLMs primarily rely on automated evaluation methods, but these methods struggle to detect implicit harmful content or produce inaccurate evaluations. Therefore, we found that existing benchmarks have low levels of harmfulness, ambiguous data, and limited diversity in image-text pair combinations. To address these issues, we propose the ELITE benchmark, a high-quality safety evaluation benchmark for VLMs, underpinned by our enhanced evaluation method, the ELITE evaluator. The ELITE evaluator explicitly incorporates a toxicity score to accurately assess harmfulness in multimodal contexts, where VLMs often provide specific, convincing, but unharmful descriptions of images. We filter out ambiguous and low-quality image-text pairs from existing benchmarks using the ELITE evaluator and generate diverse combinations of safe and unsafe image-text pairs. Our experiments demonstrate that the ELITE evaluator achieves superior alignment with human evaluations compared to prior automated methods, and the ELITE benchmark offers enhanced benchmark quality and diversity. By introducing ELITE, we pave the way for safer, more robust VLMs, contributing essential tools for evaluating and mitigating safety risks in real-world applications.
format Preprint
id arxiv_https___arxiv_org_abs_2502_04757
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ELITE: Enhanced Language-Image Toxicity Evaluation for Safety
Lee, Wonjun
Lee, Doehyeon
Choi, Eugene
Yu, Sangyoon
Yousefpour, Ashkan
Park, Haon
Ham, Bumsub
Kim, Suhyun
Computer Vision and Pattern Recognition
Computation and Language
Current Vision Language Models (VLMs) remain vulnerable to malicious prompts that induce harmful outputs. Existing safety benchmarks for VLMs primarily rely on automated evaluation methods, but these methods struggle to detect implicit harmful content or produce inaccurate evaluations. Therefore, we found that existing benchmarks have low levels of harmfulness, ambiguous data, and limited diversity in image-text pair combinations. To address these issues, we propose the ELITE benchmark, a high-quality safety evaluation benchmark for VLMs, underpinned by our enhanced evaluation method, the ELITE evaluator. The ELITE evaluator explicitly incorporates a toxicity score to accurately assess harmfulness in multimodal contexts, where VLMs often provide specific, convincing, but unharmful descriptions of images. We filter out ambiguous and low-quality image-text pairs from existing benchmarks using the ELITE evaluator and generate diverse combinations of safe and unsafe image-text pairs. Our experiments demonstrate that the ELITE evaluator achieves superior alignment with human evaluations compared to prior automated methods, and the ELITE benchmark offers enhanced benchmark quality and diversity. By introducing ELITE, we pave the way for safer, more robust VLMs, contributing essential tools for evaluating and mitigating safety risks in real-world applications.
title ELITE: Enhanced Language-Image Toxicity Evaluation for Safety
topic Computer Vision and Pattern Recognition
Computation and Language
url https://arxiv.org/abs/2502.04757