Prefill-Time Intervention for Mitigating Hallucination in Large Vision-Language Models

2026-04-28 · Chengsheng Zhang, Chenghao Sun, Xinyan Jiang, Wei Li, Xinmei Tian arxiv

Large Vision-Language Models (LVLMs) have achieved remarkable progress in visual-textual understanding, yet their reliability is critically undermined by hallucinations, i.e., the generation of factually incorrect or inconsistent responses. While recent studies using steering vectors demonstrated promise in reducing hallucinations, a notable challenge remains: they inadvertently amplify the severity of residual hallucinations. We attribute this to their exclusive focus on the decoding stage, where errors accumulate autoregressively and progressively worsen subsequent hallucinatory outputs. To address this, we propose Prefill-Time Intervention (PTI), a novel steering paradigm that intervenes only once during the prefill stage, enhancing the initial Key-Value (KV) cache before error accumulation occurs. Specifically, PTI is modality-aware, deriving distinct directions for visual and textual representations. This intervention is decoupled to steer keys toward visually-grounded objects and values to filter background noise, correcting hallucination-prone representations at their source. Extensive experiments demonstrate PTI's significant performance in mitigating hallucinations and its generalizability across diverse decoding strategies, LVLMs, and benchmarks. Moreover, PTI is orthogonal to existing decoding-stage methods, enabling plug-and-play integration and further boosting performance. Code is available at: https://github.com/huaiyi66/PTI.

📄 PDF Abstract BibTeX arXiv:2604.25642

Code (0)

등록된 구현이 없습니다.

TruthPrInt: Mitigating LVLM Object Hallucination Via Latent Truthful-Guided Pre-Intervention

2025-03-13 · Jinhao Duan, Fei Kong, Hao Cheng, James Diffenderfer 외

Object Hallucination (OH) has been acknowledged as one of the major trustworthy challenges in Large Vision-Language Models (LVLMs). Recent advancements in Large Language Models (LLMs) indicate that internal states, such …

Hallucination Object Hallucination Specificity

ICT: Image-Object Cross-Level Trusted Intervention for Mitigating Object Hallucination in Large Vision-Language Models

2024-11-22 · CVPR 2025 1 · Junzhe Chen, Tianshu Zhang, Shiyu Huang, Yuwei Niu 외

Despite the recent breakthroughs achieved by Large Vision Language Models (LVLMs) in understanding and responding to complex visual-textual contexts, their inherent hallucination tendencies limit their practical applicat…

Hallucination Object Object Hallucination

LANCET: Neural Intervention via Structural Entropy for Mitigating Faithfulness Hallucinations in LLMs

2026-01-04 · Chenxu Wang, Chaozhuo Li, Pengbo Wang, Litian Zhang 외 arxiv

Large Language Models have revolutionized information processing, yet their reliability is severely compromised by faithfulness hallucinations. While current approaches attempt to mitigate this issue through node-level a…

V-ITI: Mitigating Hallucinations in Multimodal Large Language Models via Visual Inference-Time Intervention

2025-12-03 · Nan Sun, Zhenyu Zhang, Xixun Lin, Kun Wang 외 arxiv

Multimodal Large Language Models (MLLMs) excel in numerous vision-language tasks yet suffer from hallucinations, producing content inconsistent with input visuals, that undermine reliability in precision-sensitive domain…

Enhancing Video Representations with Spatiotemporal-Semantic Residual to Mitigate Hallucinations in Video Large Multimodal Models

2026-01-30 · Yuansheng Gao, Jinman Zhao, Tong Zhang, Xingguo Xu 외 arxiv

Although Video Large Multimodal Models have achieved strong performance in video understanding, they still suffer from hallucination. Existing inference-time intervention methods usually modify videos under the contrasti…