We introduce natural adversarial examples -- real-world, unmodified, and naturally occurring examples that cause classifier accuracy to significantly degrade.
Download the natural adversarial example dataset ImageNet-A here.
Some natural adverarial examples are shown in the following figure with the correct class, predicted class, and confidence:
If you find this useful in your research, please consider citing:
@article{hendrycks2019nae,
title={Natural Adversarial Examples},
author={Dan Hendrycks and Kevin Zhao and Steven Basart and Jacob Steinhardt and Dawn Song},
journal={arXiv preprint arXiv:1907.07174},
year={2019}
}