Multi-label image classification using adaptive graph convolutional networks: From a single domain to multiple domains

• Published in: Computer vision and image understanding Vol. 247; p. 104062
• Main Authors: Singh, Inder Pal, Ghorbel, Enjie, Oyedotun, Oyebade, Aouada, Djamila
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.10.2024
• Subjects: Computer vision; Deep learning; Domain shift; Graph Convolutional Networks; Machine learning; Multi-label image classification; Unsupervised domain adaptation; Deep learning; Domain shift; Computer vision; Multi-label image classification; 41A10; Machine learning; 65D05; Unsupervised domain adaptation; 65D17; 41A05; Graph Convolutional Networks
• ISSN: 1077-3142
• DOI: 10.1016/j.cviu.2024.104062

This paper proposes an adaptive graph-based approach for multi-label image classification. Graph-based methods have been largely exploited in the field of multi-label classification, given their ability to model label correlations. Specifically, their effectiveness has been proven not only when considering a single domain but also when taking into account multiple domains. However, the topology of the used graph is not optimal as it is pre-defined heuristically. In addition, consecutive Graph Convolutional Network (GCN) aggregations tend to destroy the feature similarity. To overcome these issues, an architecture for learning the graph connectivity in an end-to-end fashion is introduced. This is done by integrating an attention-based mechanism and a similarity-preserving strategy. The proposed framework is then extended to multiple domains using an adversarial training scheme. Numerous experiments are reported on well-known single-domain and multi-domain benchmarks. The results demonstrate that our approach achieves competitive results in terms of mean Average Precision (mAP) and model size as compared to the state-of-the-art. The code will be made publicly available. •A novel graph-based approach for multi-label image classification is proposed.•It learns adaptively a graph describing label dependencies.•It is extended to cross-domain settings with an adversarial domain adaptation schema.•The proposed method achieves a good compromise between precision and network size.