Modular Graph Transformer Networks for Multi-Label Image Classification

With the recent advances in graph neural networks, there is a rising number of studies on graph-based multi-label classification with consideration object dependencies within visual data. Nevertheless, representations can become indistinguishable due to complex nature label relationships. We propose image framework based transformer networks fully exploit inter-label interactions. The paper presents modular learning scheme enhance performance by segregating computational into multiple sub-graphs modularity. proposed approach, named Modular Graph Transformer Networks (MGTN), capable employing backbones for better information propagation over different guided transformers and convolutions. validate our MS-COCO Fashion550K datasets demonstrate improvements classification. source code available at https://github.com/ReML-AI/MGTN.