MMVistaReason: Toward Open-Data and Post-Training Recipes for Multimodal Reasoning

📄 arXiv: 2610.01352v1 📥 PDF

作者: Juekai Lin, Honglin Lin, Yuqian Yuan, Xiaolong Wu, Jie Cao, Liang Liang, Yunqi Cao, Yun Zhu, Wenqiao Zhang, Lijun Wu

分类: cs.CV

发布日期: 2026-10-01

💡 一句话要点

提出MMVistaReason以解决多模态推理中的后训练挑战

🎯 匹配领域: 支柱二：RL算法与架构 (RL & Architecture) 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 多模态推理 开放数据 后训练 教师蒸馏 数据构建 模型能力 智能系统

📋 核心要点

  1. 现有的多模态推理方法在后训练阶段面临数据质量不均和监督效率低等挑战。
  2. 论文提出MMVistaReason，通过标准化开放数据、难度感知的教师蒸馏和多教师在线蒸馏等方法来提升模型性能。
  3. 实验结果表明，MVR-4B在多个基准上表现优异，且在样本使用上显著低于其他模型，展示了其有效性。

📝 摘要（中文）

开放的多模态推理模型受益于大规模推理监督，但由于数据质量不均、监督构建效率低、难度不平衡和跨领域干扰，可靠的后训练仍然具有挑战性。我们提出MMVistaReason（MVR），这是一个开放数据后训练方案，包含三个组成部分：更广泛的能力覆盖、有效的数据构建方法和专业化后整合训练。我们的分析揭示了监督难度、轨迹质量与模型能力之间的依赖关系。MVR-4B在15个多模态基准上平均得分72.8，使用的样本量比MMFineReason少约70%。

🔬 方法详解

问题定义：论文旨在解决多模态推理中后训练的可靠性问题，现有方法面临数据质量不均、监督构建效率低和跨领域干扰等痛点。

核心思路：提出MMVistaReason，通过开放数据的系统构建和能力感知的后训练策略，提升模型的推理能力和适应性。

技术框架：MVR包含三个主要模块：能力覆盖、数据构建和训练整合。首先，覆盖分析与现实推理的能力；其次，通过分阶段清洗和注释构建数据；最后，采用多教师在线蒸馏整合不同专家的能力。

关键创新：最重要的创新在于结合了难度感知的教师蒸馏与轨迹选择，形成了MVR-SFT-528K和MVR-RL-63K，显著提升了模型在多模态推理中的表现。

关键设计：在数据构建中，采用了分阶段的清洗和注释策略，并结合了基于答案可能性的轨迹选择，确保了数据的高质量和多样性。

📊 实验亮点

在15个多模态基准测试中，MVR-4B的平均得分达到72.8，超越了Qwen3.5-9B（Instruct）和MMFineReason-8B，同时使用的样本量减少约70%。当模型规模扩展至9B时，平均得分提升至74.4，超越了Qwen3.5-35B-A3B（Instruct），显示出显著的性能提升。

🎯 应用场景

该研究的潜在应用领域包括智能助手、自动驾驶、医疗影像分析等多模态任务，能够有效提升系统的推理能力和决策质量。未来，随着开放数据和后训练方法的进一步发展，MMVistaReason有望在更多实际场景中发挥重要作用。

📄 摘要（原文）

Open multimodal reasoning models have benefited from large-scale reasoning supervision, yet reliable post-training remains challenging due to uneven data quality, inefficient supervision construction, imbalanced difficulty, and cross-domain interference. We introduce MMVistaReason (MVR), an open-data post-training recipe with three components: (1) broader capability coverage across complementary Analytical and Real-World reasoning groups, emphasizing structured reasoning versus visual perception and spatial grounding; (2) efficient SFT and RL data construction, standardizing heterogeneous open data through staged cleaning and annotation, combining difficulty-aware cascaded teacher distillation with answer-likelihood-based trajectory selection to construct MVR-SFT-528K, and applying scale-specific frontier filtering for MVR-RL-63K; and (3) specialize-then-integrate training, which trains complementary RL experts and consolidates their capabilities through multi-teacher on-policy distillation (MOPD). Our analyses reveal a capacity-dependent interaction between supervision difficulty, trajectory quality, and model capacity: smaller students benefit more from selected supervision, while larger students are robust to trajectory variation and mixed-domain interference. Mixed-domain RL introduces benchmark-level negative transfer, whereas MOPD provides consistent capability integration, with the preferred KL direction varying across model scales. Across 15 multimodal benchmarks, MVR-4B achieves an average score of 72.8, outperforming Qwen3.5-9B (Instruct) and MMFineReason-8B while using about 70% fewer samples than MMFineReason. Scaling to 9B improves the average to 74.4, surpassing Qwen3.5-35B-A3B (Instruct). Overall, MMVistaReason demonstrates that systematic open-data construction and capacity-aware post-training provide a practical and scalable path toward reliable multimodal reasoning.