FactorJEPA：分解未来预测以应对拥挤城市环境
原标题：FactorJEPA: Factorizing Monolithic Futures into Layout-Agent-Interaction Channels for Crowded and Chaotic Global South Urban Worlds
AI 摘要
Hugging Face 每日论文介绍了 FactorJEPA，一种针对拥挤混乱的全球南方城市环境（称为 DENSEWORLD）的世界模型。研究团队发布了包含 22 个城市约 1000 小时视频的 DENSEWORLD-115k 数据集，并提出了 FactorJEPA 方法，将 V-JEPA 的单一未来潜在预测器分解为布局、智能体和交互三个子空间，通过可见性门控和跨因子抑制提升预测性能。实验表明，FactorJEPA 在多个指标上优于现有微调方法，且排名在 1B 和 2B 骨干上高度一致。
正文节选
FactorJEPA: Factorizing Monolithic Futures into Layout-Agent-Interaction Channels for Crowded and Chaotic Global South Urban Worlds Abstract World models have attracted significant attention for their ability to capture and predict the structure and dynamics of the physical world. In this emerging landscape, Joint Embedding Predictive Architectures (JEPA) offer a particularly compelling direction. We study a largely unexplored regime: populous, crowded, and chaotic Global South urban environment