arXiv 2503.106022026-02-08cs.CVcs.AIcs.CL

TruthPrInt: 通过潜在真实引导预干预缓解大视觉-语言模型对象幻觉

TruthPrInt: Mitigating Large Vision-Language Models Object Hallucination Via Latent Truthful-Guided Pre-Intervention

Drexel University（德雷塞尔大学）；University of Electronic Science and Technology of China（电子科技大学）；Hong Kong University of Science and Technology (Guangzhou)（香港科学与技术大学（广州））；LLNL（劳伦斯利弗莫尔国家实验室）；Lehigh University（莱斯大学）

Jinhao Duan, Fei Kong, Hao Cheng, James Diffenderfer, Bhavya Kailkhura, Lichao Sun, Xiaofeng Zhu, Xiaoshuang Shi, Kaidi Xu

备注：: 15 pages, 9 figures, the first two authors contributed equally, Accepted by ICCV 2025

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

TruthPrInt通过学习真相方向并引导解码过程，有效缓解大视觉-语言模型中的对象幻觉问题。

摘要：

对象幻觉（OH）已被认可为大视觉-语言模型（LVLMs）中的一项主要可信挑战。近年来，大型语言模型（LLMs）的进步表明，内部状态，如隐藏状态，编码生成响应的“总体真实性”。然而，内部状态在LVLMs中如何运作以及它们是否能作为“每token”幻觉指示器仍鲜有研究，这对于缓解OH至关重要。在本文中，我们首先深入探讨了存在OH问题的LVLM内部状态，并发现（1）LVLM内部状态是高特异性每token幻觉行为指示器。此外，（2）不同的LVLMs在共同的潜在子空间中编码普遍的幻觉模式，表明存在“通用真实方向”被各种LVLMs共享。基于这些发现，我们提出了Truthful-Guided Pre-Intervention（TruthPrInt），首先学习LVLM解码的真相方向，然后在LVLM解码过程中应用真相引导的推理时间干预。我们进一步提出TruthPrInt通过构建和对齐幻觉潜在子空间，增强跨LVLM和跨数据幻觉检测的可迁移性。我们在广泛的实验设置中评估了TruthPrInt，包括领域内和领域外场景，覆盖流行的LVLMs和OH基准。实验结果表明，TruthPrInt显著优于最先进的方法。代码将在https://github.com/jinhaoduan/TruthPrInt上提供。

英文摘要：

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 as hidden states, encode the "overall truthfulness" of generated responses. However, it remains under-explored how internal states in LVLMs function and whether they could serve as "per-token" hallucination indicators, which is essential for mitigating OH. In this paper, we first conduct an in-depth exploration of LVLM internal states with OH issues and discover that (1) LVLM internal states are high-specificity per-token indicators of hallucination behaviors. Moreover, (2) different LVLMs encode universal patterns of hallucinations in common latent subspaces, indicating that there exist "generic truthful directions" shared by various LVLMs. Based on these discoveries, we propose Truthful-Guided Pre-Intervention (TruthPrInt) that first learns the truthful direction of LVLM decoding and then applies truthful-guided inference-time intervention during LVLM decoding. We further propose TruthPrInt to enhance both cross-LVLM and cross-data hallucination detection transferability by constructing and aligning hallucination latent subspaces. We evaluate TruthPrInt in extensive experimental settings, including in-domain and out-of-domain scenarios, over popular LVLMs and OH benchmarks. Experimental results indicate that TruthPrInt significantly outperforms state-of-the-art methods. Codes will be available at https://github.com/jinhaoduan/TruthPrInt.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。