模仿人类认知，掌握多图推理：一种增强视觉理解的元动作框架
Mimic Human Cognition, Master Multi-Image Reasoning: A Meta-Action Framework for Enhanced Visual Understanding
把多图/多帧推理拆成 Global/Focus/Hint/Think/Answer 五个元动作，并用检索式树采样构造冷启动轨迹、再用两阶段强化学习（多样性保持探索 + DAPO 退火利用）训练，可直接迁移到需要跨图跨帧聚合证据的视觉语言任务上。
原文依据 we propose a Cognition-Inspired Meta-Action Framework (CINEMA), which decomposes multi-image reasoning into five structured meta-actions
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论多模态大语言模型（MLLMs）在单图理解上表现优异，但在多图推理场景中性能显著下降。多图推理面临图像间复杂关联以及关键信息分散于图像集合之中等根本性挑战。受人类认知过程启发，我们提出认知启发的元动作框架（CINEMA），将多图推理分解为五个结构化的元动作：全局（Global）、聚焦（Focus）、提示（Hint）、思考（Think）与回答（Answer），显式地建模人类自然采用的序列化认知步骤。针对冷启动训练，我们引入基于检索的树采样策略，生成高质量的元动作轨迹，以推理模式对模型进行引导式初始化。在强化学习阶段，我们采用两阶段范式：先是以多样性保持策略进行探索，以避免熵坍缩；随后是以 DAPO 进行退火式利用，逐步增强利用能力。为训练该模型，我们构建了一个包含 5.6 万条冷启动实例与 5.8 万条强化学习实例的数据集，覆盖多图、多帧与单图任务。我们在多图推理基准、视频理解基准与单图基准上开展了广泛评测，在多个关键基准上取得了有竞争力的最优性能。我们的模型在 MUIR 与 MVMath 基准上超越 GPT-4o，并在视频理解基准上显著优于专门的视频推理模型，验证了该人类认知启发推理框架的有效性与泛化能力。
英文原文摘要 1668 字符
While Multimodal Large Language Models (MLLMs) excel at single-image understanding, they exhibit significantly degraded performance in multi-image reasoning scenarios. Multi-image reasoning presents fundamental challenges including complex inter-relationships between images and scattered critical information across image sets. Inspired by human cognitive processes, we propose a Cognition-Inspired Meta-Action Framework (CINEMA), which decomposes multi-image reasoning into five structured meta-actions: Global, Focus, Hint, Think, and Answer, explicitly modeling the sequential cognitive steps humans naturally employ. For cold-start training, we introduce a Retrieval-Based Tree Sampling strategy that generates high-quality meta-action trajectories to bootstrap the model with reasoning patterns. During reinforcement learning, we adopt a two-stage paradigm: an exploration phase with Diversity-Preserving Strategy to avoid entropy collapse, followed by an annealed exploitation phase with DAPO to gradually strengthen exploitation. To train our model, we construct a dataset of 56k cold-start and 58k reinforcement learning instances spanning multi-image, multi-frame, and single-image tasks. We conduct extensive evaluations on multi-image reasoning benchmarks, video understanding benchmarks, and single-image benchmarks, achieving competitive state-of-the-art performance on several key benchmarks. Our model surpasses GPT-4o on the MUIR and MVMath benchmarks and notably outperforms specialized video reasoning models on video understanding benchmarks, demonstrating the effectiveness and generalizability of our human cognition-inspired reasoning framework.