ImageNet-A

Introduced by Hendrycks et al. in Natural Adversarial Examples

The ImageNet-A dataset consists of real-world, unmodified, and naturally occurring examples that are misclassified by ResNet models.

Source: On Robustness and Transferability of Convolutional Neural Networks

Homepage

Task	Dataset Variant	Best Model
Domain Generalization	ImageNet-A	Model soups
Zero-Shot Transfer Image Classification	ImageNet-A	CoCa
Prompt Engineering	ImageNet-A	POMP
Adversarial Robustness	ImageNet-A	DeiT-S
Unsupervised Domain Adaptation	ImageNet-A	EfficientNet-L2 NoisyStudent + RPL