New LA-Sign framework uses looped transformers and geometry-aware alignment to improve skeleton-based sign language recognition with multi-scale motion understanding

LA-Sign proposes a recurrent transformer approach that repeatedly refines motion understanding using shared parameters instead of stacking deeper layers

Framework captures fine-grained articulated motion across multiple spatial scales, from subtle finger movements to full body dynamics

Geometry-aware contrastive objective projects skeletal and textual features into adaptive hyperbolic space for multi-scale semantic organization

Research explores three different looping designs and multiple geometric manifolds to optimize skeleton-based isolated sign language recognition (ISLR)