CIFAR-100

Overview

The CIFAR-100 dataset is an extended version of CIFAR-10, consisting of 60,000 32×32 RGB color images across 100 fine-grained classes, grouped into 20 superclasses. Each fine class contains 600 images, and each superclass contains 3,000 images.

• Train: 50,000 images (500 per class)

• Test: 10,000 images (100 per class)

Data Structure

When accessing an example using ds[i], you will receive a dictionary with the following keys:

Key	Type	Description
image	PIL.Image.Image	32×32×3 RGB image
label	int	Fine-grained class label (0-99)
superclass	int	Coarse superclass label (0-19)

Usage Example

Basic Usage

from stable_datasets.images.cifar100 import CIFAR100

# First run will download + prepare cache, then return the split as a HF Dataset
ds = CIFAR100(split="train")

# If you omit the split (split=None), you get a DatasetDict with all available splits
ds_all = CIFAR100(split=None)

sample = ds[0]
print(sample.keys())  # {"image", "label", "superclass"}

# Optional: make it PyTorch-friendly
ds_torch = ds.with_format("torch")

Related Datasets

• CIFAR-10: Simplified version with 10 classes, sharing the same image dimensions

• CIFAR-10-C: Corrupted version of CIFAR-10 for testing robustness (available in stable-datasets as cifar10_c)

• CIFAR-100-C: Corrupted version of CIFAR-100 (available in stable-datasets as cifar100_c)

References

• Official website: https://www.cs.toronto.edu/~kriz/cifar.html

• License: MIT License

Citation

@Techreport{krizhevsky2009learning,
author = {Krizhevsky, Alex and Hinton, Geoffrey},
address = {Toronto, Ontario},
institution = {University of Toronto},
number = {0},
publisher = {Technical report, University of Toronto},
title = {Learning multiple layers of features from tiny images},
year = {2009},
title_with_no_special_chars = {Learning multiple layers of features from tiny images},
url = {https://www.cs.toronto.edu/~kriz/learning-features-2009-TR.pdf}
}