CORE: Multi-link graph attention network with inter-regional collaboration for continuous sign language recognition

Open Webpage

Yan Zhang, Wanli Xue, Yuxi Zhou, Tiantian Yuan, Shengyong Chen

Published: 2025, Last Modified: 01 Mar 2026Pattern Recognit. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•CORE: GNN-based network with ACEM (enhances visual cues) & GRCM (multi-link edges) for CSLR, improving gesture/expression analysis via spatiotemporal modeling.•ACEM: Fuses object-detected cues and grid regions via graph attention, boosting fine-grained features and reducing detection instability.•GRCM: Adds global anchor nodes and multi-links between cues to model dynamic temporal relationships, mitigating GNN over-smoothing.•CORE achieves SOTA on PHOENIX-2014(-T), CSL/CSL-Daily, proving multi-cue collaboration efficacy.