Citation

Lim, J.M. and Lim, K.M. and Goh, Pey Yun and Lee, C.P. (2026) CvTAa: Convolutional Vision Transformer with Auto-augmentation for Fine-grained Image Classification. IAENG International Journal of Computer Science, 53 (9). pp. 3629-3640. ISSN 1819656X

Text 23.pdf - Published Version Restricted to Repository staff only Download (201kB)

Abstract

Fine-grained image classification remains challenging due to subtle inter-class variations and high intra-class diversity. In this work, we propose the Convolutional Vision Transformer with Auto-augmentation (CvTAa) to address these challenges. Our approach leverages the Convolutional Vision Transformer (CvT) architecture, seamlessly combining convolutional layers with multi-head self-attention to jointly model local structural details and long-range contextual dependencies. To prevent overfitting and significantly boost model generalization, CvTAa integrates RandAugment, applying a diverse sequence of automated augmentation operations during training. The proposed CvTAa method is validated on three widely used fine-grained benchmark datasets, namely CUB-200-2011, Stanford Dogs, and Stanford Cars. Extensive ablation experiments are performed to investigate the contributions of key design choices, including learning rate scheduling, manual data augmentation, and automated data augmentation, on classification accuracy. Furthermore, CvTAa is validated through rigorous statistical significance testing, feature clustering visualization, and detailed error analysis. Ultimately, CvTAa achieves highly competitive accuracies of 80.39 on CUB-200-2011, 87.09 on Stanford Dogs, and 90.85 on Stanford Cars, demonstrating that the synergy between CvT and RandAugment provides a highly effective solution for fine-grained visual categorization.

Item Type:	Article
Uncontrolled Keywords:	Auto-augmentation, CNN-Transformer, Deep Learning
Subjects:	Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75.5-76.95 Electronic computers. Computer science
Divisions:	Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	05 Oct 2026 04:38
Last Modified:	05 Oct 2026 04:47
URII:	http://shdl.mmu.edu.my/id/eprint/16875

Downloads

Edit (login required)