Omni-Persona：系统性基准测试与改进多模态个性化
Omni-Persona: Systematic Benchmarking and Improving Omnimodal Personalization
AI总结：
Omni-Persona提出首个多模态个性化基准，通过Persona Modality Graph任务组和细粒度任务，系统分析多模态个性化中的接地行为，提出Calibrated Accuracy评估方法，揭示开源模型在音频与视觉接地上的差距及SFT与RLVR的局限性。
摘要：
尽管多模态大语言模型在文本、图像和音频方面取得进展，个性化研究仍主要集中在视觉-语言领域，缺乏统一的多模态基准测试，且缺乏方法论严谨性以应对无角色场景或系统性接地研究。我们引入Omni-Persona，首个全面的多模态个性化基准。我们将任务形式化为跨模态路由 over Persona Modality Graph，涵盖4个任务组和18个细粒度任务，共约750项。为严格诊断接地行为，我们提出Calibrated Accuracy（Cal），联合奖励正确接地和适当回避，将无角色查询纳入统一评估框架。在我们的专门实验中，三个诊断发现浮现：（i）开源模型表现出一致的音频与视觉接地差距，RLVR通过密集的规则监督部分缩小；（ii）可回答召回和参数规模是不完整的诊断，因为强召回可与无角色幻觉共存，较大模型不总是达到更高的Cal，暴露校准作为独立的评估轴；（iii）SFT受限于大规模标注地面真相监督的难度，而RLVR通过结果层面可验证反馈更一致，但在我们的奖励设计下倾向于保守行为和较低的生成质量。Omni-Persona因此成为诊断框架，揭示多模态个性化中的陷阱，指导未来训练后和奖励设计。
英文摘要：
While multimodal large language models have advanced across text, image, and audio, personalization research has remained primarily vision-language, with unified omnimodal benchmarking that jointly covers text, image, and audio still limited, and lacking the methodological rigor to account for absent-persona scenarios or systematic grounding studies. We introduce Omni-Persona, the first comprehensive benchmark for omnimodal personalization. We formalize the task as cross-modal routing over the Persona Modality Graph, encompassing 4 task groups and 18 fine-grained tasks across ~750 items. To rigorously diagnose grounding behavior, we propose Calibrated Accuracy (Cal), which jointly evaluates correct grounding and appropriate abstention, incorporating absent-persona queries within a unified evaluation framework. On our dedicated experiments, three diagnostic findings emerge: (i) recent open-weight models show a consistent audio-vs-visual grounding gap that RLVR partially narrows via dense rule-based supervision; (ii) recall score and parameter scale are incomplete diagnostics, since strong recall can coexist with absent-persona hallucination and larger models do not always achieve higher Cal, exposing calibration as a separate evaluation axis; and (iii) SFT is limited by the scalability of ground-truth annotation. RLVR improves recall but not calibrated abstention under our reward design, leaving Cal at or below the base model. Omni-Persona thus serves as a diagnostic framework that surfaces the pitfalls of omnimodal personalization, guiding future post-training and reward design.
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。
