以草稿思考：面向高效长视频理解的推测式时序推理
Thinking with Drafts: Speculative Temporal Reasoning for Efficient Long Video Understanding
把「草稿模型快速提议关键帧 + 目标模型验证并迭代精化」的推测式双模型协作引入 thinking-with-frames 流程，可直接用于降低长视频 MLLM 推理中冗余多模态上下文的开销，其帧级证据标注与 RL 训练范式也可迁移到其他需要稀疏帧采样的视频推理任务。
原文依据 a lightweight draft MLLM rapidly explores and proposes salient frames from densely sampled temporal regions
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论长视频理解对于类人智能至关重要，它使模型能够在长时序上下文中进行连贯的感知与推理。新兴的「以帧思考」范式在全局时序推理与局部帧检查之间交替进行，提升了视频多模态大语言模型（MLLM）的推理能力，但由于多模态上下文不断增长且高度冗余，该范式存在显著的效率瓶颈。为解决这一问题，我们提出 SpecTemp，一个基于强化学习的推测式时序推理框架，通过协同的双模型设计将时序感知与推理解耦。在 SpecTemp 中，轻量级草稿 MLLM 快速探索密集采样的时序区域并提出显著帧，而强大的目标 MLLM 专注于时序推理并验证草稿的提议，迭代精化其注意力直至收敛。该设计模拟了人脑的协同通路，在效率与准确率之间取得平衡。为支持训练，我们构建了 SpecTemp-80K 数据集，其特点是同时提供粗粒度证据片段与细粒度帧级证据的双层同步标注。在多个视频理解基准上的实验表明，与现有的「以帧思考」方法相比，SpecTemp 不仅保持了有竞争力的准确率，还显著加速了推理。
英文原文摘要 1449 字符
Long video understanding is essential for human-like intelligence, enabling coherent perception and reasoning over extended temporal contexts. While the emerging thinking-with-frames paradigm--which alternates between global temporal reasoning and local frame examination--has advanced the reasoning capabilities of video multi-modal large language models (MLLMs), it suffers from a significant efficiency bottleneck due to the progressively growing and redundant multi-modal context. To address this, we propose SpecTemp, a reinforcement learning-based Speculative Temporal reasoning framework that decouples temporal perception from reasoning via a cooperative dual-model design. In SpecTemp, a lightweight draft MLLM rapidly explores and proposes salient frames from densely sampled temporal regions, while a powerful target MLLM focuses on temporal reasoning and verifies the draft's proposals, iteratively refining its attention until convergence. This design mirrors the collaborative pathways of the human brain, balancing efficiency with accuracy. To support training, we construct the SpecTemp-80K dataset, featuring synchronized dual-level annotations for coarse evidence spans and fine-grained frame-level evidence. Experiments across multiple video understanding benchmarks demonstrate that SpecTemp not only maintains competitive accuracy but also significantly accelerates inference compared with existing thinking-with-frames methods.