Relation fusion propagation network for transductive few-shot learning

Yixiang Huang, Hongyu Hao, Weichao Ge, Yang Cao, Ming Wu, Chuang Zhang, Jun Guo

Abstract: Highlights•Dual-channel framework for graph-based few-shot learning, emphasizing simultaneous modeling of feature and label relations.•Relation Fusion Block (RFB) to effectively aggregate feature and label relations for improved feature propagation.•Label Shortcut Mechanism (LSM) that progressively refines sample relations with initial labels to boost semantic correlation and reduce error accumulation.•Plug-and-play capability, easily integrating with existing graph-based transductive few-shot learning methods.•Extensive experiments on four popular benchmarks, demonstrating superior performance.