Learning by Neighbor-Aware Semantics, Deciding by Open-form Flows: Towards Robust Zero-Shot Skeleton Action Recognition

The introduction of Flora, a novel method for zero-shot skeleton action recognition, represents a significant advancement in AI and computer vision. Traditional methods struggle with recognizing unseen action categories due to the lack of skeletal priors, leading to fragile alignments and rigid classifiers. Flora overcomes these challenges by employing neighbor-aware semantics for flexible semantic attunement and open-form flows for robust decision-making. By incorporating contextual cues and ensuring geometric consistency, Flora achieves stable alignment between semantic and skeleton embeddings. Extensive experiments conducted on three benchmark datasets validate the effectiveness of this approach, highlighting its potential to enhance action recognition capabilities in various applications.