arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2601.17067cs.CVcs.AI
收藏

视频生成作为世界模型的机制视角:状态与动态

A Mechanistic View on Video Generation as World Models: State and Dynamics

Hong Kong University of Science and Technology (Guangzhou)(香港理工大学(广州));Tongji University(同济大学);Kuaishou Technology(快手科技)

Luozhou Wang, Zhifei Chen, Yihua Du, Dongyu Yan, Wenhang Ge, Guibao Shen, Xinli Xu, Leyi Wu, Man Chen, Tianshuo Xu, Peiran Ren, Xin Tao, Pengfei Wan, Ying-Cong Chen

AI总结:

本文提出基于状态构建和动态建模的视频生成新分类法,旨在提升视频生成模型的物理持续性和因果推理能力。

摘要:

大规模视频生成模型已经展示了涌现的物理一致性,使其成为潜在的世界模型。然而,当前“无状态”的视频架构与经典基于状态的世界模型理论之间仍存在差距。本文通过提出以两个支柱为中心的新分类法来弥合这一差距:状态构建和动态建模。我们将状态构建分为隐式范式(上下文管理)和显式范式(潜在压缩),而动态建模则通过知识整合和架构重构进行分析。此外,我们倡导评价标准从视觉保真度转向功能基准,测试物理持续性和因果推理。我们最后指出两个关键前沿:通过数据驱动的记忆和压缩保真度来增强持续性,以及通过潜在因素解耦和推理先验整合来推进因果性。通过解决这些挑战,该领域可以从生成视觉上合理的视频发展为构建稳健、通用的世界模拟器。

英文摘要:

Large-scale video generation models have demonstrated emergent physical coherence, positioning them as potential world models. However, a gap remains between contemporary "stateless" video architectures and classic state-centric world model theories. This work bridges this gap by proposing a novel taxonomy centered on two pillars: State Construction and Dynamics Modeling. We categorize state construction into implicit paradigms (context management) and explicit paradigms (latent compression), while dynamics modeling is analyzed through knowledge integration and architectural reformulation. Furthermore, we advocate for a transition in evaluation from visual fidelity to functional benchmarks, testing physical persistence and causal reasoning. We conclude by identifying two critical frontiers: enhancing persistence via data-driven memory and compressed fidelity, and advancing causality through latent factor decoupling and reasoning-prior integration. By addressing these challenges, the field can evolve from generating visually plausible videos to building robust, general-purpose world simulators.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑