动态场景长视频生成
TLDR
重新设计时间潜在表示并采用两阶段训练，生成物体运动、相机视角和新内容一致的长期视频。
评分理由
The paper addresses a key limitation in video generation—long-term temporal consistency—by introducing a redesigned temporal latent representation and a two-phase training strategy, and it contributes new benchmark datasets. However, the abstract does not report quantitative results or comparisons, and the proposed benchmarks may not fully capture real-world complexity.
Read-first 评分解释
综合优先阅读分 32.6，由主题、引用、图谱、方法、可复现性和近期性等信号加权得到。 原始总分保留为 9。
研究版图角色
候选论文
排序敏感性
稳定性：volatile；排名波动范围：52。