1 paper · 1 filter
Zihan Su, Junhao Zhuang, Yaowei Li +10
Autoregressive video generation requires denoising the current frames while writing their key-value representations as context for future predictions. However, these two roles typi…