1 paper · 1 filter
Hao Liu, Chenghuan Huang, Ye Huang +6
Video Diffusion Transformers process long spatio-temporal sequences, making self-attention the main bottleneck in high-resolution video generation. Training-free sparse attention r…