In the field of image recognition, the scale and diversity of datasets are crucial for model training. This study proposes a novel cross-validation dataset pruning method with data balancing (CVDP-DB). The method evaluates the correctness of training samples and their predicted probabilities, scoring them and subsequently performing precise pruning. We focus on pruning hard samples that are incorrectly predicted and easy samples that are correctly predicted with high probability, while retaining samples with high prediction uncertainty. These samples form a refined coreset for training models. This approach not only optimizes the feature distribution of the dataset but also enhances the model’s ability to recognize key samples. Experimental results show that the CVDP-DB method demonstrates excellent classification performance across various models and datasets. Notably, the applicability of CVDP-DB transcends the constraints of specific models or datasets, and the method surpasses state-of-the-art (SOTA) technologies.

Generative models have emerged as powerful tools capable of generating photorealistic images, spawning a wide range of applications across various domains. However, effectively integrating generative models into image classification tasks remains an open problem. Our analysis reveals that current generative data augmentation methods, as well as traditional data augmentation techniques, have limitations in simultaneously ensuring both fidelity (faithful foreground) and diversity (rich background contexts). To address this challenge, we propose Decomposition-Recomposition Data Augmentation (DRMix), an innovative intra-class data augmentation method. DRMix decomposes images into foreground-background and foreground parts, then performs diversified background recomposition and intra-class foreground recomposition, achieving dual diversity enhancement at both the image and part levels, and strikes a better trade-off between fidelity and diversity. Experimental results demonstrate that DRMix significantly improves performance across multiple tasks, including image classification, few-shot learning, and weakly-supervised object localization (WSOL).