Large Vision-Language Models (LVLMs) can reason from image-text inputs and perform well in various multimodal tasks. Despite this success, they are affected by language priors and often produce hallucinations. Hallucinations denote generated content that is grammatically and syntactically coherent, yet bears no match or direct relevance to visual input. To address this problem, we propose Residual Decoding (ResDec). It is a novel training-free method that uses historical information to aid decoding. The method relies on the internal implicit reasoning mechanism and token logits evolution mechanism of LVLMs to correct biases. Extensive experiments demonstrate that ResDec effectively suppresses hallucinations induced by language priors, significantly improves visual grounding, and reduces object hallucinations. In addition to mitigating hallucinations, ResDec also performs exceptionally well on comprehensive LVLM benchmarks, highlighting its broad applicability.
@article{arxiv.2602.01047,
title = {Residual Decoding: Mitigating Hallucinations in Large Vision-Language Models via History-Aware Residual Guidance},
author = {Xinrong Chen and Xu Chu and Yingmin Qiu and Hengyuan Zhang and Jing Xiong and Shiyu Tang and Shuai Liu and Shaokang Yang and Cheng Yang and Hayden Kwok-Hay So and Ngai Wong},
journal= {arXiv preprint arXiv:2602.01047},
year = {2026}
}