MIHBench: Benchmarking and Mitigating Multi-Image Hallucinations in Multimodal Large Language Models

2025-08-01

AI Summary

Prior work predominantly addresses hallucination in single-image understanding, leaving multi-image hallucination—particularly concerning object existence, count accuracy, and cross-view identity consistency—largely unexplored and unbenchmarked. Method: We introduce MIHBench, the first dedicated benchmark for multi-image object hallucination, comprising three semantic reasoning tasks. Leveraging MIHBench, we systematically analyze key factors influencing hallucination—including image count, per-image hallucination propensity, proportion of same-object images, and negative sample placement—and propose a dynamic attention balancing mechanism within a Transformer architecture. This mechanism jointly models semantic consistency and performs count-aware reasoning. Contribution/Results: Evaluated on multiple state-of-the-art multimodal large language models, our approach significantly reduces hallucination rates, enhances cross-image semantic integration, and improves reasoning robustness. MIHBench establishes a reproducible evaluation framework and an optimization paradigm for multi-image understanding.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Despite growing interest in hallucination in Multimodal Large Language Models, existing studies primarily focus on single-image settings, leaving hallucination in multi-image scenarios largely unexplored. To address this gap, we conduct the first systematic study of hallucinations in multi-image MLLMs and propose MIHBench, a benchmark specifically tailored for evaluating object-related hallucinations across multiple images. MIHBench comprises three core tasks: Multi-Image Object Existence Hallucination, Multi-Image Object Count Hallucination, and Object Identity Consistency Hallucination, targeting semantic understanding across object existence, quantity reasoning, and cross-view identity consistency. Through extensive evaluation, we identify key factors associated with the occurrence of multi-image hallucinations, including: a progressive relationship between the number of image inputs and the likelihood of hallucination occurrences; a strong correlation between single-image hallucination tendencies and those observed in multi-image contexts; and the influence of same-object image ratios and the positional placement of negative samples within image sequences on the occurrence of object identity consistency hallucination. To address these challenges, we propose a Dynamic Attention Balancing mechanism that adjusts inter-image attention distributions while preserving the overall visual attention proportion. Experiments across multiple state-of-the-art MLLMs demonstrate that our method effectively reduces hallucination occurrences and enhances semantic integration and reasoning stability in multi-image scenarios.

Problem

• Study multi-image hallucinations in multimodal large language models
• Benchmark object-related hallucinations across multiple images
• Mitigate hallucinations via Dynamic Attention Balancing mechanism

Innovation

• Dynamic Attention Balancing mechanism adjusts attention
• MIHBench evaluates multi-image object hallucinations
• Analyzes image input number and hallucination likelihood