CutPaste&Find: Efficient Multimodal Hallucination Detector with Visual-aid Knowledge Base

TL;DR

CutPaste&Find is a lightweight, training-free multimodal hallucination detection framework that uses a visual-aid knowledge base and off-the-shelf modules to efficiently identify non-existent objects in LVLM outputs.

Contribution

It introduces a novel, efficient, and training-free detection method leveraging a visual-aid knowledge base, reducing reliance on expensive API calls and iterative validation.

Findings

Achieves competitive detection performance on benchmark datasets.

Significantly more efficient and cost-effective than existing methods.

Operates without requiring LVLM inference or training.

Abstract

Large Vision-Language Models (LVLMs) have demonstrated impressive multimodal reasoning capabilities, but they remain susceptible to hallucination, particularly object hallucination where non-existent objects or incorrect attributes are fabricated in generated descriptions. Existing detection methods achieve strong performance but rely heavily on expensive API calls and iterative LVLM-based validation, making them impractical for large-scale or offline use. To address these limitations, we propose CutPaste\&Find, a lightweight and training-free framework for detecting hallucinations in LVLM-generated outputs. Our approach leverages off-the-shelf visual and linguistic modules to perform multi-step verification efficiently without requiring LVLM inference. At the core of our framework is a Visual-aid Knowledge Base that encodes rich entity-attribute relationships and associated image representations. We introduce a scaling factor to refine similarity scores, mitigating the issue of suboptimal alignment values even for ground-truth image-text pairs. Comprehensive evaluations on benchmark datasets, including POPE and R-Bench, demonstrate that CutPaste\&Find achieves competitive hallucination detection performance while being significantly more efficient and cost-effective than previous methods.