arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2611.06433cs.CV
收藏

DreamRealm:探索与实现幻觉世界以构建可信的视觉-语言模型

DreamRealm: Exploring and Materializing Hallucination Worlds for Trustworthy Vision-Language Models

Alibaba Group(阿里巴巴集团)

Benlei Cui, Chunyang Chai, Yiliang Zhang, Xiaoqian Xia, Chengwen Yao, Song Li, Longtao Huang, Yuwen Zhai, Haiwen Hong

AI总结:

DreamRealm是一个后训练框架,通过挖掘Qwen3.5-9B的幻觉构建RealmSet,结合RealmMatch训练目标,在6个基准上显著提升了视觉-语言模型的幻觉抑制与感知性能。

摘要:

多模态大语言模型(MLLMs)在正确响应依赖于细微视觉细节时仍易出现视觉幻觉。一类后训练方法通过图像检索、生成或编辑构建视觉对比以提升细粒度感知,但这些对比并非围绕模型自身的幻觉构建,难以确保监督针对其实际感知混淆。另一类方法在固定图像条件下通过响应校正或偏好学习抑制幻觉输出,但对图像输入侧监督信号的设计探索不足。受人类纠正误解的方式启发,我们提出DreamRealm,这是一个后训练框架,用于探索多样的潜在幻觉响应、最小程度编辑原始图像使其有效,并训练模型为每个响应分配对应的支持图像。我们的数据构建流水线RealmForgePipe从目标模型中挖掘观察到的和潜在的幻觉,通过图像编辑将其实现。使用从Qwen3.5-9B中挖掘的幻觉,我们构建了已实现幻觉数据集(RealmSet)。我们的训练目标RealmMatch构建图像-响应组合的似然得分矩阵,结合监督学习与双向对比损失以强化正确匹配并抑制不匹配。与原始Qwen3.5-9B模型相比,DreamRealm在6个视觉感知与幻觉基准上提升了性能,包括HallusionBench准确率提升7.26个百分点、MMVP配对准确率提升11.34个百分点。结合对数据构建与训练方法的消融研究,这些结果表明幻觉分配是提升视觉感知并缓解幻觉的有前景范式。

英文摘要:

Multimodal large language models (MLLMs) remain prone to visual hallucinations when correct responses depend on subtle visual details. One line of post-training methods constructs visual contrasts through image retrieval, generation, or editing to improve fine-grained perception. However, these contrasts are not built around the model's own hallucinations, making it difficult to ensure that supervision targets its actual perceptual confusions. Another line suppresses hallucinated outputs through response correction or preference learning under fixed image conditions, but leaves the design of supervision signals on the image-input side underexplored. Inspired by how humans correct misconceptions, we introduce DreamRealm, a post-training framework that explores diverse latent hallucinated responses, minimally edits original images to make them valid, and trains the model to assign each response to its supporting image. Our data construction pipeline, RealmForgePipe, mines observed and latent hallucinations from a target model and materializes them through image editing. Using hallucinations mined from Qwen3.5-9B, we construct the Realized Hallucination Dataset (RealmSet). Our training objective, RealmMatch, constructs a likelihood score matrix across image--response combinations and combines supervised learning with bidirectional contrastive losses to reinforce correct matches and suppress mismatches. Compared with the original Qwen3.5-9B model, DreamRealm improves performance across six visual perception and hallucination benchmarks, including gains of 7.26 percentage points in HallusionBench accuracy and 11.34 percentage points in MMVP pair accuracy. Together with ablation studies on data construction and training methods, these results demonstrate that hallucination assignment is a promising paradigm for improving visual perception and mitigating hallucinations.

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。

↑