The MetaShift dataset contains 12,868 sets of natural images covering 410 categories, which can be used to measure and evaluate the robustness of machine learning models to data shifts.

Citation

@InProceedings{liang2022metashift, title={MetaShift: A Dataset of Datasets for Evaluating Contextual Distribution Shifts and Training Conflicts}, author={Weixin Liang and James Zou}, booktitle={International Conference on Learning Representations}, year={2022}, url={https://openreview.net/forum?id=MTex8qKavoS} }