ImageNet-C
An image robustness benchmark dataset introduced by Hendrycks et al. in 2019. It applies 15 types of synthetic corruptions—including noise, blur, and weather degradations—to the ImageNet validation set, with each corruption at five severity levels, yielding 75 subsets in total. It is widely used to evaluate model robustness under distribution shift.
Timeline (last 90 days)
The performance degradation on real weather is not visible on the synthetic held-out test set, because there is a systematic gap between the distribution of synthetic corruptions and that of real degradations.
ImageNet-C was proposed by Hendrycks et al. in 2019; it applies 15 types of synthetic corruptions to the ImageNet validation set, each with 5 severity levels, forming 75 subsets in total, and its corruptions are image-space post-processing operations rather than being based on real physical imaging processes.
All Facts (2)
The performance degradation on real weather is not visible on the synthetic held-out test set, because there is a systematic gap between the distribution of synthetic corruptions and that of real degradations.
50%UnverifiedImageNet-C was proposed by Hendrycks et al. in 2019; it applies 15 types of synthetic corruptions to the ImageNet validation set, each with 5 severity levels, forming 75 subsets in total, and its corruptions are image-space post-processing operations rather than being based on real physical imaging processes.
50%