PhysInOne：一体化视觉物理学习与推理套件
PhysInOne: Visual Physics Learning and Reasoning in One Suite
为物理感知视频生成与世界模型提供带3D几何、语义、运动、物理属性与文本标注的大规模合成视频数据，可直接用于微调视频生成/预测基础模型并评测其物理合理性。
原文依据 PhysInOne provides 2 million videos across 153,810 dynamic 3D scenes
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论我们提出 PhysInOne，一个大规模合成数据集，旨在缓解 AI 系统在物理基础训练数据上的严重稀缺。与仅包含数百或数千个样本的现有数据集不同，PhysInOne 提供 200 万段视频，覆盖 153,810 个动态 3D 场景，涵盖力学、光学、流体动力学和磁学中的 71 种基本物理现象。与以往工作不同，我们的场景具有复杂背景下的多物体交互，并提供全面的真值标注，包括 3D 几何、语义、动态运动、物理属性和文本描述。我们展示了 PhysInOne 在四个新兴应用上的有效性：物理感知视频生成、长/短期未来帧预测、物理属性估计和运动迁移。实验表明，在 PhysInOne 上微调基础模型可显著提升物理合理性，同时也暴露出在建模复杂物理动力学和估计内在属性方面的关键差距。作为同类中规模最大的数据集，PhysInOne 在数量级上超越先前工作，为推进生成、仿真和具身 AI 中的物理基础世界模型建立了新的基准。
英文原文摘要 1236 字符
We present PhysInOne, a large-scale synthetic dataset addressing the critical scarcity of physically-grounded training data for AI systems. Unlike existing datasets limited to merely hundreds or thousands of examples, PhysInOne provides 2 million videos across 153,810 dynamic 3D scenes, covering 71 basic physical phenomena in mechanics, optics, fluid dynamics, and magnetism. Distinct from previous works, our scenes feature multiobject interactions against complex backgrounds, with comprehensive ground-truth annotations including 3D geometry, semantics, dynamic motion, physical properties, and text descriptions. We demonstrate PhysInOne's efficacy across four emerging applications: physics-aware video generation, long-/short-term future frame prediction, physical property estimation, and motion transfer. Experiments show that fine-tuning foundation models on PhysInOne significantly enhances physical plausibility, while also exposing critical gaps in modeling complex physical dynamics and estimating intrinsic properties. As the largest dataset of its kind, orders of magnitude beyond prior works, PhysInOne establishes a new benchmark for advancing physics-grounded world models in generation, simulation, and embodied AI.