从上下文到线索的缩放:面向多图像 MLLMs 的分层偏好优化
Zooming from Context to Cue: Hierarchical Preference Optimization for Multi-Image MLLMs
AI总结:
针对多模态大语言模型在多图像理解中的跨模态错位与幻觉问题,提出多级偏好优化框架 CcDPO,通过上下文级与细针级优化增强单图像感知,并构建数据集 MultiScope-42k,显著减少幻觉并提升任务性能。
摘要:
多模态大语言模型(MLLMs)擅长单图像任务,但在多图像理解上存在困难,这归因于跨模态错位,从而导致幻觉(上下文遗漏、混淆和误解)。使用直接偏好优化(DPO)的现有方法将优化限制在输入序列中的单一图像参考内,忽略了整体上下文建模。我们提出了上下文到线索直接偏好优化(CcDPO),这是一个多级偏好优化框架,通过将视觉线索从序列上下文缩放至局部细节,来增强多图像设置中的单图像感知。其特点包括:(i) 上下文级优化:重新评估 MLLMs 多图像上下文理解背后的认知偏差,并整合一系列低成本的全局序列偏好以缓解偏差。(ii) 细针级优化:通过区域定向的视觉提示和多模态偏好监督,将注意力引向细粒度的视觉细节。为支持可扩展优化,我们还构建了 MultiScope-42k,这是一个自动生成的数据集,包含高质量的多级偏好对。实验表明,CcDPO 显著减少了幻觉,并在一般的单图像和多图像任务中产生了一致的性能提升。
英文摘要:
Multi-modal Large Language Models (MLLMs) excel at single-image tasks but struggle with multi-image understanding due to cross-modal misalignment, leading to hallucinations (context omission, conflation, and misinterpretation). Existing methods using Direct Preference Optimization (DPO) constrain optimization to a solitary image reference within the input sequence, neglecting holistic context modeling. We propose Context-to-Cue Direct Preference Optimization (CcDPO), a multi-level preference optimization framework that enhances per-image perception in multi-image settings by zooming into visual clues -- from sequential context to local details. It features: (i) Context-Level Optimization : Re-evaluates cognitive biases underlying MLLMs' multi-image context comprehension and integrates a spectrum of low-cost global sequence preferences for bias mitigation. (ii) Needle-Level Optimization : Directs attention to fine-grained visual details through region-targeted visual prompts and multimodal preference supervision. To support scalable optimization, we also construct MultiScope-42k, an automatically generated dataset with high-quality multi-level preference pairs. Experiments show that CcDPO significantly reduces hallucinations and yields consistent performance gains across general single- and multi-image tasks.
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。