English(EN)Latent Spatial Memory for Video World Models
新基准和框架推动视频世界建模发展
作者PulseAugur 编辑部·[13 个来源]·
研究人员推出了“ImageTime”，这是一个旨在评估图像生成模型理解和表示时间变化能力的新基准。该基准通过要求模型生成一个动作的四个有序关键状态来评估时空一致性，超越了单一图像质量指标。此外，还开发了一个名为BiWM的新框架，利用双向自回归来推进开源交互式视频世界模型，旨在提高生成质量和推理速度。另一篇论文提出了一种用于视频世界模型的“潜在空间记忆”，将场景信息直接存储在扩散潜在空间中，从而显著加快生成速度并减小内存占用。
AI
arXiv:2603.11395v3 Announce Type: replace-cross Abstract: Continual reinforcement learning challenges agents to acquire new skills while retaining previously learned ones with the goal of improving performance in both past and future tasks. Most existing approaches rely on model-…
We present MoVerse, a real-time video world model that creates an interactively navigable scene from a single narrow-field-of-view image. This setting is challenging because the input observes only a small fraction of the environment, while interactive roaming requires a complete…
MoVerse generates real-time interactive video from single images by creating 360° panoramas and 3D Gaussian scaffolds, enabling efficient rendering through diffusion-based techniques.
arXiv:2606.10620v1 Announce Type: cross Abstract: Image generation models now produce high-quality static images, yet their ability to represent how a visual world changes over time remains poorly understood. Practical workflows such as storyboarding, step-by-step illustration, r…
arXiv:2606.10135v1 Announce Type: cross Abstract: Transitioning bidirectional video diffusion models into an autoregressive paradigm improves the interactivity of video world models, but existing causal pipelines need many stages (control fine-tuning, autoregressive training, cau…
Image generation models now produce high-quality static images, yet their ability to represent how a visual world changes over time remains poorly understood. Practical workflows such as storyboarding, step-by-step illustration, reference-guided editing, and video previsualizatio…
arXiv cs.AI
TIER_1English(EN)·Jewon Yeom, Hanseul Kim, Jeongjae Park, Sungmok Jung, Jaejin Lee, Taesup Kim·
arXiv:2606.07687v1 Announce Type: cross Abstract: Video world models are increasingly used to provide predictive visual representations, yet it remains unclear which pretraining signals induce action-relevant structure in their latent spaces. We study this question through a unif…
Video world models that maintain 3D spatial consistency across generated frames typically rely on explicit point cloud memory constructed in RGB space. This design is both computationally expensive, requiring repeated rendering and VAE encoding, and inherently lossy, as the round…
Latent spatial memory for video world models stores 3D scene information directly in diffusion latent space, eliminating pixel-space reconstruction overhead and achieving faster generation with reduced memory usage.
arXiv:2606.13376v1 Announce Type: new Abstract: We present MoVerse, a real-time video world model that creates an interactively navigable scene from a single narrow-field-of-view image. This setting is challenging because the input observes only a small fraction of the environmen…
We present MoVerse, a real-time video world model that creates an interactively navigable scene from a single narrow-field-of-view image. This setting is challenging because the input observes only a small fraction of the environment, while interactive roaming requires a complete…
arXiv:2606.09828v1 Announce Type: new Abstract: Video world models that maintain 3D spatial consistency across generated frames typically rely on explicit point cloud memory constructed in RGB space. This design is both computationally expensive, requiring repeated rendering and …
Video world models that maintain 3D spatial consistency across generated frames typically rely on explicit point cloud memory constructed in RGB space. This design is both computationally expensive, requiring repeated rendering and VAE encoding, and inherently lossy, as the round…
