MKZuziak/cifar10h

---
dataset_info:
features:
- name: image
dtype: image
- name: label
dtype:
class_label:
names:
'0': airplane
'1': automobile
'2': bird
'3': cat
'4': deer
'5': dog
'6': frog
'7': horse
'8': ship
'9': truck
- name: expert_probs
list: float32
length: 10
- name: expert_counts
list: int32
length: 10
- name: expert_argmax
list: int32
length: 10
splits:
- name: train
num_bytes: 23931580
num_examples: 10000
download_size: 24078418
dataset_size: 23931580
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
license: mit
task_categories:
- image-classification
tags:
- CIFAR10
- CIFAR10H
- L2D
- LearningtoDefer
- LearningtoAsk
- HumanLabels
pretty_name: CIFAR10H
size_categories:
- 1K<n<10K
---

# CIFAR-10H Hugging Face Dataset

This repository contains a Hugging Face dataset build of CIFAR-10H, an extension of the CIFAR-10 test set with human-annotated label distributions.

## Dataset Description

CIFAR-10H adds human uncertainty information to the CIFAR-10 test images by providing:
- `expert_probs`: probability distributions over the 10 CIFAR-10 classes
- `expert_counts`: raw human vote counts for each class
- `expert_argmax`: one-hot encoded labels from the human-majority choice

The dataset is constructed from the original CIFAR-10 test images together with the CIFAR-10H annotation files.

## Original Citation

Please credit the original CIFAR-10H authors when using this dataset:

Peterson, Joshua C.; Battleday, Ruairidh M.; Griffiths, Thomas L.; Russakovsky, Olga.
"Human uncertainty makes classification more robust." arXiv preprint arXiv:1908.07086, 2019.

Source repository:
https://github.com/jcpeterson/cifar-10h

This HuggingFace conversion was done Zuziak, Maciej K.