REVAL: A Comprehension Evaluation on Reliability and Values of Large Vision-Language Models

2025-03-20

AI Summary

Existing LVLM evaluation benchmarks suffer from narrow coverage, failing to systematically assess model reliability (e.g., factual consistency, adversarial robustness) and value alignment (e.g., ethics, safety, privacy). To address this, we propose ReliVLM—the first dual-axis comprehensive evaluation framework for LVLMs—accompanied by a large-scale VQA benchmark of 144K samples spanning hallucination, adversarial vulnerability, bias, toxicity, and privacy leakage. Our methodology introduces multi-granularity question construction, adversarial image perturbations, moral scenario reasoning, and privacy-sensitivity testing, coupled with a standardized scoring system. Extensive evaluation across 26 state-of-the-art LVLMs reveals strong performance in perception and toxicity mitigation, yet significant deficiencies persist in adversarial robustness, privacy preservation, and ethical reasoning—highlighting critical gaps in current LVLM development.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

The rapid evolution of Large Vision-Language Models (LVLMs) has highlighted the necessity for comprehensive evaluation frameworks that assess these models across diverse dimensions. While existing benchmarks focus on specific aspects such as perceptual abilities, cognitive capabilities, and safety against adversarial attacks, they often lack the breadth and depth required to provide a holistic understanding of LVLMs' strengths and limitations. To address this gap, we introduce REVAL, a comprehensive benchmark designed to evaluate the extbf{RE}liability and extbf{VAL}ue of LVLMs. REVAL encompasses over 144K image-text Visual Question Answering (VQA) samples, structured into two primary sections: Reliability, which assesses truthfulness (eg, perceptual accuracy and hallucination tendencies) and robustness (eg, resilience to adversarial attacks, typographic attacks, and image corruption), and Values, which evaluates ethical concerns (eg, bias and moral understanding), safety issues (eg, toxicity and jailbreak vulnerabilities), and privacy problems (eg, privacy awareness and privacy leakage). We evaluate 26 models, including mainstream open-source LVLMs and prominent closed-source models like GPT-4o and Gemini-1.5-Pro. Our findings reveal that while current LVLMs excel in perceptual tasks and toxicity avoidance, they exhibit significant vulnerabilities in adversarial scenarios, privacy preservation, and ethical reasoning. These insights underscore critical areas for future improvements, guiding the development of more secure, reliable, and ethically aligned LVLMs. REVAL provides a robust framework for researchers to systematically assess and compare LVLMs, fostering advancements in the field.

Problem

• Assessing reliability and values of Large Vision-Language Models comprehensively
• Evaluating LVLMs on truthfulness, robustness, ethics, safety, and privacy
• Identifying vulnerabilities in adversarial scenarios and ethical reasoning

Innovation

• Comprehensive benchmark for LVLM evaluation
• 144K VQA samples for reliability and values
• Assesses 26 models including GPT-4o