VIGIL: Tackling Hallucination Detection in Image Recontextualization

TL;DR

VIGIL introduces a detailed benchmark and detection framework for hallucinations in multimodal image recontextualization, categorizing errors into five types to improve understanding and evaluation of large multimodal models.

Contribution

It provides the first fine-grained categorization and detection pipeline for hallucinations in multimodal image recontextualization tasks, filling a significant gap in model evaluation.

Findings

Effective multi-stage detection pipeline demonstrated

Comprehensive categorization of hallucination types

Open-source release of dataset and tools

Abstract

We introduce VIGIL (Visual Inconsistency & Generative In-context Lucidity), the first benchmark dataset and framework providing a fine-grained categorization of hallucinations in the multimodal image recontextualization task for large multimodal models (LMMs). While existing research often treats hallucinations as a uniform issue, our work addresses a significant gap in multimodal evaluation by decomposing these errors into five categories: pasted object hallucinations, background hallucinations, object omission, positional & logical inconsistencies, and physical law violations. To address these complexities, we propose a multi-stage detection pipeline. Our architecture processes recontextualized images through a series of specialized steps targeting object-level fidelity, background consistency, and omission detection, leveraging a coordinated ensemble of open-source models, whose effectiveness is demonstrated through extensive experimental evaluations. Our approach enables a deeper understanding of where the models fail with an explanation; thus, we fill a gap in the field, as no prior methods offer such categorization and decomposition for this task. To promote transparency and further exploration, we openly release VIGIL, along with the detection pipeline and benchmark code, through our GitHub repository: https://github.com/mlubneuskaya/vigil and Data repository: https://huggingface.co/datasets/joannaww/VIGIL.