Class-Adaptive Data Augmentation for Image Classification

Data augmentation is a widely used regularization technique for improving the performance of convolutional neural networks (CNNs) in image classification tasks. To improve effectiveness data augmentation, it important to find label-preserving transformations that fit domain knowledge given dataset. In several real-world datasets, appropriate policies differ between classes, owing their different characteristics. this paper, we propose class-adaptive method utilizes class-specific policies. First, train CNN without augmentation. Subsequently, derive suitable policy each class through an optimization procedure maximize degree transformation while maintaining property CNNs. Finally, re-train model using based on derived Through experiments benchmark datasets with constraints, demonstrate proposed achieves comparable or higher accuracy than baseline methods same all classes. Additionally, confirm are consistent