ImageNet-C

This repository contains the ImageNet-C dataset from Benchmarking Neural Network Robustness to Common Corruptions and Perturbations.

noise.tar (21GB) contains gaussian_noise, shot_noise, and impulse_noise.
blur.tar (7GB) contains defocus_blur, glass_blur, motion_blur, and zoom_blur.
weather.tar (12GB) contains frost, snow, fog, and brightness.
digital.tar (7GB) contains contrast, elastic_transform, pixelate, and jpeg_compression.
extra.tar (15GB) contains speckle_noise, spatter, gaussian_blur, and saturate.

AlexNet Errors used to normalize the Corruption Errors are as follows,
Gaussian Noise: 88.6%,
Shot Noise: 89.4%,
Impulse Noise: 92.3%,
Defocus Blur: 82.0%,
Glass Blur: 82.6%,
Motion Blur: 78.6%,
Zoom Blur: 79.8%,
Snow: 86.7%,
Frost: 82.7%,
Fog: 81.9%,
Brightness: 56.5%,
Contrast: 85.3%,
Elastic Transformation: 64.6%,
Pixelate: 71.8%,
JPEG: 60.7%,
Speckle Noise: 84.5%,
Gaussian Blur: 78.7%,
Spatter: 71.8%,
Saturate: 65.8%

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

    @article{hendrycks2019robustness,
      title={Benchmarking Neural Network Robustness to Common Corruptions and Perturbations},
      author={Hendrycks, Dan and Dietterich, Thomas},
      journal={Proceedings of the International Conference on Learning Representations},
      year={2019}
    }