Multitask Art Painting Categorization by Convolution and Transformer Hybrid Architecture

Automatic categorization of fine art paintings across multiple semantic facets, such as artist, style, and genre, is fundamental for large-scale digital archiving, semantic indexing, and knowledge organization of cultural heritage collections. In this paper, we propose convolutional neural network (CNN)-Transformer Hybrid Attention model for art paintings categorization (CTHArt), a CNN-Transformer Hybrid Attention network for multitask art painting categorization. The model employs a dual-branch hybrid backbone that combines a CNN stream for fine-grained local texture modeling and a Transformer stream for global compositional and stylistic context learning. To further exploit inter-facet semantic dependencies, we introduce a Cross-Task Attention Head, which enables task-specific classifiers to exchange information through learnable cross-attention interactions. This design supports coordinated facet prediction consistent with knowledge organization principles. We evaluate the proposed framework on three benchmark datasets. Experimental results demonstrate that CTHArt consistently achieves state-of-the-art performance. The proposed approach provides an effective and scalable solution for artificial intelligence (AI)-assisted knowledge organization of art collections.