生成，但要验证：通过回顾性重采样减少视觉语言模型中的幻觉
Generate, but Verify: Reducing Hallucination in Vision-Language Models with Retrospective Resampling
AI总结：
本文提出 REVERSE 框架，结合幻觉感知训练与推理时自验证回顾性重采样，使 VLM 能检测并动态修正视觉幻觉，在 CHAIR-MSCOCO 和 HaloQuest 上分别最高提升12%和34%。
摘要：
视觉语言模型（VLM）擅长视觉理解，但经常出现视觉幻觉，即生成对不存在的对象、动作或概念的描述，这在安全关键应用中构成重大风险。现有的幻觉缓解方法通常遵循两种范式之一：生成调整，通过修改解码行为使文本与视觉输入对齐；事后验证，由外部模型评估并纠正输出。尽管有效，但生成调整方法往往依赖启发式规则且缺乏纠正机制，而事后验证较为复杂，通常需要多个模型，并且倾向于拒绝输出而非改进输出。在本工作中，我们提出 REVERSE，这是一个将幻觉感知训练与即时自验证相结合的统一框架。通过利用一个包含超过130万个半合成样本的新型幻觉验证数据集，以及一种新颖的推理时回顾性重采样技术，我们的方法使 VLM 能够在生成过程中检测幻觉，并动态修正这些幻觉。评估表明，REVERSE 实现了最先进的幻觉减少效果，在 CHAIR-MSCOCO 上较现有最佳方法最高提升12%，在 HaloQuest 上最高提升34%。数据集、模型和代码可在 https://reverse-vlm.github.io 获取。
英文摘要：
Vision-Language Models (VLMs) excel at visual understanding but often suffer from visual hallucinations, where they generate descriptions of nonexistent objects, actions, or concepts, posing significant risks in safety-critical applications. Existing hallucination mitigation methods typically follow one of two paradigms: generation adjustment, which modifies decoding behavior to align text with visual inputs, and post-hoc verification, where external models assess and correct outputs. While effective, generation adjustment methods often rely on heuristics and lack correction mechanisms, while post-hoc verification is complicated, typically requiring multiple models and tending to reject outputs rather than refine them. In this work, we introduce REVERSE, a unified framework that integrates hallucination-aware training with on-the-fly self-verification. By leveraging a new hallucination-verification dataset containing over 1.3M semi-synthetic samples, along with a novel inference-time retrospective resampling technique, our approach enables VLMs to both detect hallucinations during generation and dynamically revise those hallucinations. Our evaluations show that REVERSE achieves state-of-the-art hallucination reduction, outperforming the best existing methods by up to 12% on CHAIR-MSCOCO and 34% on HaloQuest. Our dataset, model, and code are available at: https://reverse-vlm.github.io.
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。
