Knowledge-based visual question classification using quaternion hypergraph consistent network

doi:10.1016/j.ipm.2025.104591

Knowledge-based visual question classification using quaternion hypergraph consistent network

2026-01-13

0

PRE

AI

DOI：10.1016/j.ipm.2025.104591

Original

Original request for help

Save

Abstract

En 中文

• A quaternion hypergraph consistent network is proposed to integrate three fine-grained properties. • Independence loss is designed to make textual features and visual features maintain modality-specific and distinct from each other. • Knowledge consistency loss is proposed to constrain the textual-knowledge features and visual-knowledge features. • Experimental results demonstrate that our method achieves better performance comparable to existing mainstream approaches.

Knowledge-based visual question classification using quaternion hypergraph consistent network

Abstract

Journal

Organization