MotionHiFlow: Text-to-motion via hierarchical flow matching
MotionHiFlow is a hierarchical flow matching framework that generates 3D human motions progressively from low to high temporal scales. It uses a novel cross-scale transition process and integrates a Text-Motion Diffusion Transformer and a topology-aware Motion VAE. This combination allows the model to achieve superior semantic alignment and fine-grained motion details on benchmarks like HumanML3D.