Blind Spot Navigation: Evolutionary Discovery of Sensitive Semantic Concepts for LVLMs

TL;DR

This paper explores the sensitive semantic concepts that cause large vision-language models to hallucinate or fail, introducing a novel evolutionary framework that combines LLMs and T2I models to identify these concepts.

Contribution

It presents the first systematic method to discover sensitive semantics in LVLMs using an evolutionary approach with LLMs and T2I models, revealing new insights into model vulnerabilities.

Findings

LVLMs are susceptible to hallucinations triggered by specific semantic concepts.

The proposed method effectively identifies sensitive semantics across multiple LVLMs.

Insights into LVLM vulnerabilities can guide future robustness improvements.

Abstract

Adversarial attacks aim to generate malicious inputs that mislead deep models, but beyond causing model failure, they cannot provide certain interpretable information such as ``\textit{What content in inputs make models more likely to fail?}'' However, this information is crucial for researchers to specifically improve model robustness. Recent research suggests that models may be particularly sensitive to certain semantics in visual inputs (such as ``wet,'' ``foggy''), making them prone to errors. Inspired by this, in this paper we conducted the first exploration on large vision-language models (LVLMs) and found that LVLMs indeed are susceptible to hallucinations and various errors when facing specific semantic concepts in images. To efficiently search for these sensitive concepts, we integrated large language models (LLMs) and text-to-image (T2I) models to propose a novel semantic evolution framework. Randomly initialized semantic concepts undergo LLM-based crossover and mutation operations to form image descriptions, which are then converted by T2I models into visual inputs for LVLMs. The task-specific performance of LVLMs on each input is quantified as fitness scores for the involved semantics and serves as reward signals to further guide LLMs in exploring concepts that induce LVLMs. Extensive experiments on seven mainstream LVLMs and two multimodal tasks demonstrate the effectiveness of our method. Additionally, we provide interesting findings about the sensitive semantics of LVLMs, aiming to inspire further in-depth research.