Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Xiaofeng, Zhu, Yuanchao, Gu, Chaochen, Yuan, Xiaosong, Zhao, Qiyan, Cao, Jiawei, Tang, Feilong, Fan, Sinan, Shen, Yaomin, Shen, Chen, Tang, Hao
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2601.20279
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910003252690944
author Zhang, Xiaofeng
Zhu, Yuanchao
Gu, Chaochen
Yuan, Xiaosong
Zhao, Qiyan
Cao, Jiawei
Tang, Feilong
Fan, Sinan
Shen, Yaomin
Shen, Chen
Tang, Hao
author_facet Zhang, Xiaofeng
Zhu, Yuanchao
Gu, Chaochen
Yuan, Xiaosong
Zhao, Qiyan
Cao, Jiawei
Tang, Feilong
Fan, Sinan
Shen, Yaomin
Shen, Chen
Tang, Hao
contents Recent studies have examined attention dynamics in large vision-language models (LVLMs) to detect hallucinations. However, existing approaches remain limited in reliably distinguishing hallucinated from factually grounded outputs, as they rely solely on forward-pass attention patterns and neglect gradient-based signals that reveal how token influence propagates through the network. To bridge this gap, we introduce LVLMs-Saliency, a gradient-aware diagnostic framework that quantifies the visual grounding strength of each output token by fusing attention weights with their input gradients. Our analysis uncovers a decisive pattern: hallucinations frequently arise when preceding output tokens exhibit low saliency toward the prediction of the next token, signaling a breakdown in contextual memory retention. Leveraging this insight, we propose a dual-mechanism inference-time framework to mitigate hallucinations: (1) Saliency-Guided Rejection Sampling (SGRS), which dynamically filters candidate tokens during autoregressive decoding by rejecting those whose saliency falls below a context-adaptive threshold, thereby preventing coherence-breaking tokens from entering the output sequence; and (2) Local Coherence Reinforcement (LocoRE), a lightweight, plug-and-play module that strengthens attention from the current token to its most recent predecessors, actively counteracting the contextual forgetting behavior identified by LVLMs-Saliency. Extensive experiments across multiple LVLMs demonstrate that our method significantly reduces hallucination rates while preserving fluency and task performance, offering a robust and interpretable solution for enhancing model reliability. Code is available at: https://github.com/zhangbaijin/LVLMs-Saliency
format Preprint
id arxiv_https___arxiv_org_abs_2601_20279
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Hallucination Begins Where Saliency Drops
Zhang, Xiaofeng
Zhu, Yuanchao
Gu, Chaochen
Yuan, Xiaosong
Zhao, Qiyan
Cao, Jiawei
Tang, Feilong
Fan, Sinan
Shen, Yaomin
Shen, Chen
Tang, Hao
Computer Vision and Pattern Recognition
Recent studies have examined attention dynamics in large vision-language models (LVLMs) to detect hallucinations. However, existing approaches remain limited in reliably distinguishing hallucinated from factually grounded outputs, as they rely solely on forward-pass attention patterns and neglect gradient-based signals that reveal how token influence propagates through the network. To bridge this gap, we introduce LVLMs-Saliency, a gradient-aware diagnostic framework that quantifies the visual grounding strength of each output token by fusing attention weights with their input gradients. Our analysis uncovers a decisive pattern: hallucinations frequently arise when preceding output tokens exhibit low saliency toward the prediction of the next token, signaling a breakdown in contextual memory retention. Leveraging this insight, we propose a dual-mechanism inference-time framework to mitigate hallucinations: (1) Saliency-Guided Rejection Sampling (SGRS), which dynamically filters candidate tokens during autoregressive decoding by rejecting those whose saliency falls below a context-adaptive threshold, thereby preventing coherence-breaking tokens from entering the output sequence; and (2) Local Coherence Reinforcement (LocoRE), a lightweight, plug-and-play module that strengthens attention from the current token to its most recent predecessors, actively counteracting the contextual forgetting behavior identified by LVLMs-Saliency. Extensive experiments across multiple LVLMs demonstrate that our method significantly reduces hallucination rates while preserving fluency and task performance, offering a robust and interpretable solution for enhancing model reliability. Code is available at: https://github.com/zhangbaijin/LVLMs-Saliency
title Hallucination Begins Where Saliency Drops
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2601.20279