Yiteng Xu

Achieve realistic and responsive avatar interactions without expensive full-sequence inference, thanks to a novel modular framework that adapts pre-trained motion priors to diverse interaction scenarios.