ChainMPQ: Interleaved Text-Image Reasoning Chains for Mitigating Relation Hallucinations

Researchers propose ChainMPQ, a training-free method to reduce relation hallucinations in Large Vision-Language Models (LVLMs) by using interleaved text-image reasoning chains. The approach addresses the most common but least studied type of AI hallucination by sequentially analyzing subjects, objects, and their relationships through multi-perspective questioning.

• →ChainMPQ is a training-free method that significantly reduces relation hallucinations in Large Vision-Language Models without requiring additional model training.
• →Relation hallucinations account for the largest proportion of LVLM errors but have received the least research attention compared to object and attribute hallucinations.
• →The method uses multi-perspective questioning to analyze three core relationship components: subject, object, and the connecting relation.
• →ChainMPQ creates interleaved chains of images and text where earlier textual and visual memories support subsequent reasoning steps.
• →Experimental results across multiple LVLMs and benchmarks demonstrate substantial improvements in relational inference accuracy.