Tiny ImageNetV2, -R, -A, MedMNIST-C, EuroSAT-C
收藏资源简介:
本研究提出了多个针对小规模数据集的测试集变体,包括Tiny ImageNetV2, -R, -A,以及医疗和航空领域的MedMNIST-C和EuroSAT-C。这些数据集通过将ImageNet的泛化和鲁棒性基准转移到小规模数据领域来创建,旨在评估和提升Vision Transformers在有限数据情况下的性能和鲁棒性。数据集的创建过程涉及将真实数据与生成数据结合,以扩展现有的小规模数据集。这些数据集主要应用于图像分类任务,特别是在医疗影像和航空影像分析领域,以解决数据稀缺和模型鲁棒性不足的问题。
This study proposes several test set variants for small-scale datasets, including Tiny ImageNetV2, -R, -A, as well as MedMNIST-C and EuroSAT-C from the medical and aviation domains. These datasets are developed by transferring the generalization and robustness benchmarks of ImageNet to the small-scale data domain, aiming to evaluate and improve the performance and robustness of Vision Transformers under limited data conditions. The dataset creation process combines real-world data with generated data to expand existing small-scale datasets. These datasets are mainly applied to image classification tasks, especially in medical imaging and aerial image analysis, to address the issues of data scarcity and insufficient model robustness.




