KA-GCN: Kernel-Attentive Graph Convolutional Network for 3D face analysis

Download PDF
Open Webpage

Francesco Agnelli, Giuseppe Facchi, Giuliano Grossi, Raffaella Lanzarotti

Published: 01 Jan 2025, Last Modified: 09 Nov 2025Array 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•The KA-Conv integrates kernel and attention mechanisms to optimize graph structures.•Two kernel distance measures prioritize nodes based on geometrical proximity.•Ablation studies show KA-GCN outperforms variants, balancing close and distant nodes.•KA-GCN is tested on multiple 3D face analysis tasks, outperforming competitors.