MMFakeBench: A Mixed-Source Multimodal Misinformation Detection Benchmark for LVLMs

TL;DR

This paper introduces MMFakeBench, a comprehensive benchmark for mixed-source multimodal misinformation detection, evaluates existing methods and LVLMs, and proposes MMD-Agent to improve detection accuracy in realistic scenarios.

Contribution

It presents the first benchmark for mixed-source MMD, evaluates multiple models on this benchmark, and introduces MMD-Agent to enhance detection performance.

Findings

Current methods perform poorly on mixed-source MMD

LVLMs struggle under zero-shot settings for misinformation detection

MMD-Agent significantly improves detection accuracy and generalization

Abstract

Current multimodal misinformation detection (MMD) methods often assume a single source and type of forgery for each sample, which is insufficient for real-world scenarios where multiple forgery sources coexist. The lack of a benchmark for mixed-source misinformation has hindered progress in this field. To address this, we introduce MMFakeBench, the first comprehensive benchmark for mixed-source MMD. MMFakeBench includes 3 critical sources: textual veracity distortion, visual veracity distortion, and cross-modal consistency distortion, along with 12 sub-categories of misinformation forgery types. We further conduct an extensive evaluation of 6 prevalent detection methods and 15 Large Vision-Language Models (LVLMs) on MMFakeBench under a zero-shot setting. The results indicate that current methods struggle under this challenging and realistic mixed-source MMD setting. Additionally, we propose MMD-Agent, a novel approach to integrate the reasoning, action, and tool-use capabilities of LVLM agents, significantly enhancing accuracy and generalization. We believe this study will catalyze future research into more realistic mixed-source multimodal misinformation and provide a fair evaluation of misinformation detection methods.