A Comprehensive Analysis for Visual Object Hallucination in Large Vision-Language Models

TL;DR

This paper investigates the causes of visual object hallucination in large vision-language models, analyzing model components and proposing mitigation strategies, supported by new benchmarks for evaluating hallucination issues.

Contribution

It provides a comprehensive analysis of hallucination sources in LVLMs and introduces two benchmarks to evaluate and mitigate these hallucinations.

Findings

Identified key sources of hallucination in model components

Proposed targeted mitigation methods for each component

Developed two benchmarks for hallucination evaluation

Abstract

Large Vision-Language Models (LVLMs) demonstrate remarkable capabilities in multimodal tasks, but visual object hallucination remains a persistent issue. It refers to scenarios where models generate inaccurate visual object-related information based on the query input, potentially leading to misinformation and concerns about safety and reliability. Previous works focus on the evaluation and mitigation of visual hallucinations, but the underlying causes have not been comprehensively investigated. In this paper, we analyze each component of LLaVA-like LVLMs -- the large language model, the vision backbone, and the projector -- to identify potential sources of error and their impact. Based on our observations, we propose methods to mitigate hallucination for each problematic component. Additionally, we developed two hallucination benchmarks: QA-VisualGenome, which emphasizes attribute and relation hallucinations, and QA-FB15k, which focuses on cognition-based hallucinations.