CAI: Caption-Sensitive Attention Intervention for Mitigating Object Hallucination in Large Vision-Language Models

TL;DR

This paper introduces CAI, a training-free method that uses attention patterns to reduce object hallucination in large vision-language models, improving accuracy with minimal inference overhead.

Contribution

The paper proposes a novel, plug-and-play attention intervention method that mitigates hallucination without additional training or significant inference costs.

Findings

Achieves state-of-the-art hallucination mitigation across multiple benchmarks.

Effective for both discriminative and generative vision-language tasks.

Minimal additional inference cost required.

Abstract

Although Large Vision-Language Models (LVLMs) have demonstrated powerful capabilities in interpreting visual information, they frequently produce content that deviates from visual information, leading to object hallucination. To tackle this, recent works mostly depend on expensive manual annotations and training cost, or significantly increase inference time. In this work, we observe that LVLMs' attention to visual information is significantly stronger when answering caption queries compared to non-caption queries. Inspired by this phenomenon, we propose Caption-sensitive Attention Intervention (CAI), a training-free, plug-and-play hallucination mitigation method that leverages the attention activation pattern in response to caption queries to enhance LVLMs' visual perception capability. Extensive experimental results across four benchmarks covering both discriminative and generative tasks, demonstrate that CAI achieves state-of-the-art (SOTA) hallucination mitigating performance only with minimal additional inference cost.