Skip to content

Latest commit

 

History

History
20 lines (13 loc) · 930 Bytes

README.md

File metadata and controls

20 lines (13 loc) · 930 Bytes

Rethinking Natural Adversarial Examples for Classification Models

We introduce ImageNet-A-Plus, a new natural adversarial examples dataset. The new dataset can be used to study the robustness of classification models to the internal variance of objects without considering the background disturbance.

Download the natural adversarial example dataset ImageNet-A-Plus here.

Some examples in ImageNet-A-Plus are shown below:

Citation

If you find this useful in your research, please consider citing:

@article{xiaoli2021nae,
  title={Rethinking Natural Adversarial Examples for Classification Models},
  author={Xiao Li and Jianmin Li and Ting Dai and Jie Shi and Jun Zhu and Xiaolin Hu},
  journal={arXiv preprint arXiv:2102.11731},
  year={2021}
}