Dual Adversarial Graph Neural Networks for Multi-label Cross-modal Retrieval

Cross-modal retrieval has become an active study field with the expanding scale of multimodal data. To date, most existing methods transform data into a common representation space where semantic similarities between items can be directly measured across different modalities. However, these typically suffer from following limitations: 1) They usually attempt to bridge modality gap by designing losses in which may not sufficient eliminate potential heterogeneity modalities space. 2) treat labels as independent individuals and ignore label relationships are important for constructing links In this work, we propose novel Dual Adversarial Graph Neural Networks (DAGNN) composed dual generative adversarial networks multi-hop graph neural networks, learn modality-invariant discriminative representations cross-modal retrieval. Firstly, construct project Secondly, leverage layer aggregation mechanism is proposed exploit propagation information, capture correlation dependency inter-dependent classifiers. Comprehensive experiments conducted on two benchmark datasets, NUS-WIDE MIRFlickr, indicate superiority DAGNN.