Graph Neural Networks for Semi-Supervised Image Classification with Multi-Feature Aggregation

The article presents a novel approach for employing Graph Neural Networks (GNNs), specifically Graph Convolutional Networks (GCNs), in semi-supervised image classification by integrating multiple feature and graph representations from various extractors. Experimental results indicate that combining these diverse representations and applying rank aggregation techniques significantly improves classification accuracy, particularly in scenarios with limited labeled data. This work is relevant for practitioners as it highlights effective strategies to enhance GNN performance using manifold learning and diverse feature extraction methods, which can be critical for real-world applications with sparse labeled datasets.