Decoupled and boosted learning for skeleton-based dynamic hand gesture recognition

Yangke Li, Guangshun Wei, Christian Desrosiers, Yuanfeng Zhou

Abstract: Highlights•A lightweight dual-stream boosted learning framework for hand gesture recognition.•The TMB-Stream boosts significant motion features via the hybrid attention module.•The SSB-Stream uses the multi-scale auto-encoder to boost spatial latent features.•The SSB-Stream utilizes the multi-scale fusion module to fuse multi-scale features.•An exhaustive experimental study on three well-known available datasets.