ReactionMamba: Generating Short & Long Human Reaction Sequences
This addresses the challenge of generating realistic and diverse long human motion sequences for applications like animation or virtual reality, representing an incremental advancement over existing methods.
The paper tackles the problem of generating long 3D human reaction motions, such as dance and martial arts, by proposing ReactionMamba, a framework that integrates a motion VAE with Mamba-based state-space models. The result is competitive performance in realism, diversity, and long-sequence generation on three datasets, with substantial improvements in inference speed compared to previous methods.
We present ReactionMamba, a novel framework for generating long 3D human reaction motions. Reaction-Mamba integrates a motion VAE for efficient motion encoding with Mamba-based state-space models to decode temporally consistent reactions. This design enables ReactionMamba to generate both short sequences of simple motions and long sequences of complex motions, such as dance and martial arts. We evaluate ReactionMamba on three datasets--NTU120-AS, Lindy Hop, and InterX--and demonstrate competitive performance in terms of realism, diversity, and long-sequence generation compared to previous methods, including InterFormer, ReMoS, and Ready-to-React, while achieving substantial improvements in inference speed.