robro/cifar10-c-parquet

Dataset Card for CIFAR10-C

This dataset is simply an update of the original dataset into the parquet format which should work with the current (circa 2025) huggingface dataset library

Dataset Details

Dataset Description

The CIFAR-10-C dataset is an extension of CIFAR-10 designed to evaluate model robustness to common corruptions. It consists of 950,000 images derived from the original CIFAR-10 test set (10,000 images) by applying 19 different corruption types at 5 severity levels. The corruptions include noise, blur, weather effects, and digital distortions. This dataset is widely used for benchmarking robustness in image classification tasks.

Dataset Sources

• Homepage: https://github.com/hendrycks/robustness
• Paper: Hendrycks, D., & Dietterich, T. (2019). Benchmarking neural network robustness to common corruptions and perturbations. arXiv preprint arXiv:1903.12261.

Dataset Structure

Each sample in the dataset contains:

• image: A 32×32 RGB image in PNG format

• label: An integer between 0 and 9, representing the class

• corruption_name: The name of the applied corruption

• corruption_level: An integer between 1 and 5 indicating severity

Total images: 950,000

Classes: 10 (airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck)

Corruptions: 19 types (e.g., Gaussian noise, motion blur, contrast, fog, frost, elastic transform, pixelate, JPEG compression, etc.)

Severity Levels: 5 (ranging from least to most severe)

Splits:

• Train: 950,000 images

Image specs: PNG format, 32×32 pixels, RGB

Example Usage

Below is a quick example of how to load this dataset via the Hugging Face Datasets library.

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("robro/cifar10-c-parquet", split="train", trust_remote_code=False)
classes = ["airplane","automobile","bird","cat","deer","dog","frog","horse","ship","truck",]

# Access a sample from the dataset
example = dataset[0]
image = example["image"]
label = example["label"]

image.show()  # Display the image
print(f"Label: {classes[label]}")

Citation

BibTeX:

@article{hendrycks2019benchmarking,
  title={Benchmarking neural network robustness to common corruptions and perturbations},
  author={Hendrycks, Dan and Dietterich, Thomas},
  journal={arXiv preprint arXiv:1903.12261},
  year={2019}
}