ImageNet-RIB

ImageNet-RIB（Robustness Inheritance Benchmark）是由麻省理工学院的研究团队创建的一个新的基准数据集，旨在评估经过微调的模型在多样化下游任务和评估OOD数据集上的鲁棒性。该数据集包含8个与ImageNet相关的OOD数据集，用于微调和评估模型的鲁棒性。数据集的创建过程涉及对预训练模型在不同下游数据集上的微调，并在剩余的OOD数据集上进行评估。ImageNet-RIB的应用领域主要集中在机器学习模型的鲁棒性评估和微调策略的开发，旨在解决模型在面对分布外样本时的性能下降问题。

ImageNet-RIB (Robustness Inheritance Benchmark) is a novel benchmark dataset developed by a research team at the Massachusetts Institute of Technology (MIT). It is designed to evaluate the robustness of fine-tuned models across diverse downstream tasks and out-of-distribution (OOD) datasets. This dataset includes 8 ImageNet-related OOD datasets, which are used for fine-tuning models and assessing their robustness. The construction of ImageNet-RIB involves fine-tuning pre-trained models on different downstream datasets, followed by evaluation on the remaining OOD datasets. The main application scenarios of ImageNet-RIB focus on robustness evaluation of machine learning models and development of fine-tuning strategies, aiming to address the performance degradation issue of models when encountering out-of-distribution samples.