• Description:

ImageNet-A is a set of images labelled with ImageNet labels that were obtained by collecting new data and keeping only those images that ResNet-50 models fail to correctly classify. For more details please refer to the paper.

The label space is the same as that of ImageNet2012. Each example is represented as a dictionary with the following keys:

Split Examples
'test' 7,500
  • Feature structure:
    'file_name': Text(shape=(), dtype=string),
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=1000),
  • Feature documentation:
Feature Class Shape Dtype Description
file_name Text string
image Image (None, None, 3) uint8
label ClassLabel int64


  • Citation:
  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},