1 paper · 1 filter
Weijia Dou, Hui Li, Jiahao Cui +3
Streaming video generation models typically rely on temporal-centric memory, which organizes historical context as raw frames, chunk segments, or unclustered tokens. This organizat…