用视频思考：面向长视频推理的多模态工具增强强化学习
Thinking With Videos: Multimodal Tool-Augmented Reinforcement Learning for Long Video Reasoning
把「按需采样视频帧」做成可调用的视觉工具箱，并用 DGRPO 平衡多任务 RL 中的难度差异，为长视频 MLLM 推理提供了一条可复用的 agentic 训练范式；做视频问答或时序定位时可直接借鉴其多任务数据构造与 RL 策略。
原文依据 the model can densely sample new video frames on demand and generate multimodal CoT for precise long video reasoning
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（MLLM）的视频推理能力对视频问答、时序定位等下游任务至关重要。已有工作探索了基于文本的思维链（CoT）推理，但这类方法往往跨模态交互有限、幻觉增多，在视频较长或推理链较长时尤为明显。为应对这些挑战，我们提出 VITAL（Video Intelligence via Tool-Augmented Learning），一种全新的端到端智能体式视频推理框架。借助视觉工具箱，模型可按需密集采样新的视频帧，并生成多模态思维链，从而实现精确的长视频推理。我们观察到，时序定位与问答对视频理解任务相互促进。因此，我们构建了两个高质量的多任务视频推理数据集：用于监督微调的 MTVR-CoT-72k 和用于强化学习的 MTVR-RL-110k。此外，我们提出难度感知的组相对策略优化算法（DGRPO），以缓解多任务强化学习中的难度不平衡问题。在十一个具有挑战性的视频理解基准上的大量实验表明，VITAL 具备先进的推理能力，在视频问答与时序定位任务上优于现有方法，在长视频场景中尤为突出。代码已发布于 https://github.com/zhang9302002/ThinkingWithVideos。
英文原文摘要 1450 字符
The video reasoning ability of multimodal large language models (MLLMs) is crucial for downstream tasks like video question answering and temporal grounding. While recent approaches have explored text-based chain-of-thought (CoT) reasoning for MLLMs, these methods often suffer from limited cross-modal interaction and increased hallucination, especially with longer videos or reasoning chains. To address these challenges, we propose Video Intelligence via Tool-Augmented Learning (VITAL), a novel end-to-end agentic video reasoning framework. With a visual toolbox, the model can densely sample new video frames on demand and generate multimodal CoT for precise long video reasoning. We observe that temporal grounding and question answering are mutually beneficial for video understanding tasks. Therefore, we construct two high-quality multi-task video reasoning datasets MTVR-CoT-72k for supervised fine-tuning and MTVR-RL-110k for reinforcement learning. Moreover, we propose a Difficulty-aware Group Relative Policy Optimization algorithm (DGRPO) to mitigate difficulty imbalance in multi-task reinforcement learning. Extensive experiments on eleven challenging video understanding benchmarks demonstrate the advanced reasoning ability of VITAL, outperforming existing methods in video question answering and temporal grounding tasks, especially in long video scenarios. Code is available at https://github.com/zhang9302002/ThinkingWithVideos.