WorldReel：具有一致几何与运动建模的 4D 视频生成
WorldReel: 4D Video Generation with Consistent Geometry and Motion Modeling
把点图、相机轨迹和稠密光流作为显式 4D 表示与 RGB 帧联合生成，为视频生成提供了一条用几何监督约束时空一致性的可复用路线；做视频扩散或世界模型的研究者可以直接借鉴其 4D 监督信号构造与合成/真实数据混合训练策略。
原文依据 WorldReel jointly produces RGB frames together with 4D scene representations, including pointmaps, camera trajectory, and dense flow mapping
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论近期的视频生成器已能实现惊人的照片级真实感，但在 3D 层面仍从根本上不一致。本文提出 WorldReel，一种原生具备时空一致性的 4D 视频生成器。WorldReel 在生成 RGB 帧的同时联合生成 4D 场景表示，包括点图、相机轨迹与稠密光流映射，从而在时间维度上实现连贯的几何与外观建模。我们的显式 4D 表示强制约束出一个在跨视角与动态内容下持续存在的单一底层场景，使得视频即便在大幅非刚性运动和显著相机移动下仍保持一致。我们通过精心组合合成数据与真实数据来训练 WorldReel：合成数据提供精确的 4D 监督（几何、运动与相机），真实视频则贡献视觉多样性与真实感。这种混合使 WorldReel 能够泛化到野外拍摄的素材，同时保持较强的几何保真度。大量实验表明，WorldReel 在动态场景与移动相机下的一致性视频生成上达到了新的最优水平，在几何一致性、运动连贯性指标上有所提升，并相较竞争方法减少了视角间伪影。我们相信 WorldReel 使视频生成更接近 4D 一致的世界建模，即智能体能够通过单一且稳定的时空表示对场景进行渲染、交互与推理。项目主页：https://bshfang.github.io/worldreel/
英文原文摘要 1458 字符
Recent video generators achieve striking photorealism, yet remain fundamentally inconsistent in 3D. We present WorldReel, a 4D video generator that is natively spatio-temporally consistent. WorldReel jointly produces RGB frames together with 4D scene representations, including pointmaps, camera trajectory, and dense flow mapping, enabling coherent geometry and appearance modeling over time. Our explicit 4D representation enforces a single underlying scene that persists across viewpoints and dynamic content, yielding videos that remain consistent even under large non-rigid motion and significant camera movement.We train WorldReel by carefully combining synthetic and real data: synthetic data provides precise 4D supervision (geometry, motion, and camera), while real videos contribute visual diversity and realism. This blend allows WorldReel to generalize to in-the-wild footage while preserving strong geometric fidelity. Extensive experiments demonstrate that WorldReel sets a new state-of-the-art for consistent video generation with dynamic scenes and moving cameras, improving metrics of geometric consistency, motion coherence, and reducing view-time artifacts over competing methods. We believe that WorldReel brings video generation closer to 4D-consistent world modeling, where agents can render, interact, and reason about scenes through a single and stable spatiotemporal representation. Project page: https://bshfang.github.io/worldreel/