1 paper
Xinjian Wu, Hongmei Wang, Yuan Zhou +1
The scalability of high-fidelity video diffusion models (VDMs) is constrained by two key sources of redundancy: the quadratic complexity of global spatio-temporal attention and the…