The Role of Background Information in Reducing Object Hallucination in Vision-Language Models: Insights from Cutoff API Prompting

TL;DR

This paper investigates how background information influences the reduction of object hallucination in vision-language models, emphasizing the importance of background context preservation for improving model reliability.

Contribution

It provides new insights into the role of background context in visual prompting, highlighting its significance in mitigating hallucinations in VLMs.

Findings

Background preservation reduces hallucinations

Attention-driven prompting effectiveness depends on context

Background context is crucial for reliable VLM outputs

Abstract

Vision-Language Models (VLMs) occasionally generate outputs that contradict input images, constraining their reliability in real-world applications. While visual prompting is reported to suppress hallucinations by augmenting prompts with relevant area inside an image, the effectiveness in terms of the area remains uncertain. This study analyzes success and failure cases of Attention-driven visual prompting in object hallucination, revealing that preserving background context is crucial for mitigating object hallucination.