datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
cifar10
Dataset Card for CIFAR-10
Dataset Summary
The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images per class. There are 50000 training images and 10000 test images.
The dataset is divided into five training batches and one test batch, each with 10000 images. The test batch contains exactly 1000 randomly-selected images from each class. The training batches contain the remaining images in random order, but some training batches may contain… See the full description on the dataset page: https://huggingface.co/datasets/uoft-cs/cifar10.cifar100
Dataset Card for CIFAR-100
Dataset Summary
The CIFAR-100 dataset consists of 60000 32x32 colour images in 100 classes, with 600 images
per class. There are 500 training images and 100 testing images per class. There are 50000 training images and 10000 test images. The 100 classes are grouped into 20 superclasses.
There are two labels per image - fine label (actual class) and coarse label (superclass).
Supported Tasks and Leaderboards
image-classification: The… See the full description on the dataset page: https://huggingface.co/datasets/uoft-cs/cifar100.cifar100-enrichedThe CIFAR-100 dataset consists of 60000 32x32 colour images in 100 classes, with 600 images
per class. There are 500 training images and 100 testing images per class. There are 50000 training images and 10000 test images. The 100 classes are grouped into 20 superclasses.
There are two labels per image - fine label (actual class) and coarse label (superclass).cifar10
Dataset Card for CIFAR-10
Dataset Summary
The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images per class. There are 50000 training images and 10000 test images.
The dataset is divided into five training batches and one test batch, each with 10000 images. The test batch contains exactly 1000 randomly-selected images from each class. The training batches contain the remaining images in random order, but some training batches may contain… See the full description on the dataset page: https://huggingface.co/datasets/mteb/cifar10.TTA-CIFAR-100-C
TTA-CIFAR-100-C
Mirror of CIFAR-100-C (Hendrycks & Dietterich, ICLR 2019) with a
revision pin for reproducible test-time adaptation evaluation.
Upstream: Zenodo record 3555552
License: CC BY 4.0 (matches upstream)
Sibling: TTA-CIFAR-10-C
Maintained as part of: TTA-Evaluation-Harness
Citation
@inproceedings{hendrycks2019benchmarking,
title={Benchmarking Neural Network Robustness to Common Corruptions and Perturbations},
author={Hendrycks, Dan and Dietterich… See the full description on the dataset page: https://huggingface.co/datasets/WNJXYK/TTA-CIFAR-100-C.cifar100-lt
Dataset Card for CIFAR-100-LT (Long Tail)
Dataset Summary
Note (March 2026): This dataset has been migrated from a Python loading script to
parquet format for compatibility with datasets v4.4+. No trust_remote_code=True
is needed. Available configs: r-10, r-20, r-50, r-100.
The CIFAR-100-LT imbalanced dataset is comprised of under 60,000 color images, each measuring 32x32 pixels,
distributed across 100 distinct classes.
The number of samples within each class… See the full description on the dataset page: https://huggingface.co/datasets/tomas-gajarsky/cifar100-lt.TTA-CIFAR-10-C
TTA-CIFAR-10-C
Mirror of CIFAR-10-C (Hendrycks & Dietterich, ICLR 2019) with a revision
pin for reproducible test-time adaptation evaluation.
Upstream: Zenodo record 2535967
License: CC BY 4.0 (matches upstream)
SHA256 of upstream tarball: c72763e101c723b7c507b96205f7e938912a5d587376173b825850cf3cb876a7
Maintained as part of: TTA-Evaluation-Harness
Citation
@inproceedings{hendrycks2019benchmarking,
title={Benchmarking Neural Network Robustness to Common… See the full description on the dataset page: https://huggingface.co/datasets/WNJXYK/TTA-CIFAR-10-C.cifar10-lt
Dataset Card for CIFAR-10-LT (Long Tail)
Dataset Summary
Note (March 2026): This dataset has been migrated from a Python loading script to
parquet format for compatibility with datasets v4.4+. No trust_remote_code=True
is needed. Available configs: r-10, r-20, r-50, r-100.
The CIFAR-10-LT imbalanced dataset is comprised of under 60,000 color images, each measuring 32x32 pixels,
distributed across 10 distinct classes.
The number of samples within each class decreases… See the full description on the dataset page: https://huggingface.co/datasets/tomas-gajarsky/cifar10-lt.MICO-CIFAR10
MICO CIFAR-10 challenge dataset
Mico Argentatus (Silvery Marmoset) - William Warby/Flickr
For the accompanying code, visit the GitHub repository of the competition: https://github.com/microsoft/MICO/.
Getting Started
The starting kit notebook for this task is available at: https://github.com/microsoft/MICO/tree/main/starting-kit.
In the starting kit notebook you will find a walk-through of how to load the data and make your first submission.
We also provide a… See the full description on the dataset page: https://huggingface.co/datasets/szanella/MICO-CIFAR10.cifar10
Dataset Specifications
Contains the entire CIFAR10 dataset, downloaded via PyTorch, then split and saved as .png files representing 32x32 images.
There a three splits, perfectly balanced class-wise:
train: 49,000 out of the original 50,000 samples from the training set of CIFAR10;
calibration: 1,000 left-out samples from the training set;
test: 10,000 samples, the entire original test set.
File Structure
Files are archives <split>/<classname>.zip. Each… See the full description on the dataset page: https://huggingface.co/datasets/ego-thales/cifar10.CIFAR10
🖼️ CIFAR10 (Extracted from PyTorch Vision)
The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images per class. There are 50000 training images and 10000 test images.
ℹ️ Dataset Details
📖 Dataset Description
The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images per class. There are 50000 training images and 10000 test images. The classes are completely mutually exclusive. There is no… See the full description on the dataset page: https://huggingface.co/datasets/p2pfl/CIFAR10.cifar10-enrichedThe CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images
per class. There are 50000 training images and 10000 test images.
This version if CIFAR-10 is enriched with several metadata such as embeddings, baseline results and label error scores.CIFAR-10
CIFAR-10 - Object Recognition in Images
Benchmark dataset for object classification.🖼️ 60,000 32x32 color images🏷️ 10 classes📁 Format: PNG, CSV📦 Files: 4🧪 Subset of the 80 million tiny images dataset
Dataset Summary
CIFAR-10 is a widely used computer vision dataset consisting of 60,000 32x32 color images in 10 mutually exclusive classes. It was created by Alex Krizhevsky, Vinod Nair, and Geoffrey Hinton. The dataset is a labeled subset of the 80 million tiny… See the full description on the dataset page: https://huggingface.co/datasets/KDKCE/CIFAR-10.CIFAR-CThe license is to the original authors (see below)!
This repository contains the CIFAR-10-C dataset from Benchmarking Neural Network Robustness to Common Corruptions and Perturbations. We are currently hosting it on Hugging Face due to an increased latency from Zenodo.
We are not the original authors. If you find this useful in your research, please consider citing:
@article{hendrycks2019robustness,
title={Benchmarking Neural Network Robustness to Common Corruptions and Perturbations}… See the full description on the dataset page: https://huggingface.co/datasets/torch-uncertainty/CIFAR-C.BTL3_CIFAR-10
CIFAR-10 Feature Representations (BTL3)
This dataset contains pre-extracted feature embeddings from the CIFAR-10 dataset, produced using several pretrained image classification models.The goal is to enable fast experimentation, classifier prototyping, and model comparison without needing to train or forward pass large models in Colab.
Dataset Source
The original CIFAR-10 dataset is MIT-licensed and available here:https://www.cs.toronto.edu/~kriz/cifar.html
This dataset… See the full description on the dataset page: https://huggingface.co/datasets/LeTienDat/BTL3_CIFAR-10.cifar100
Dataset Card for CIFAR-100
Dataset Summary
The CIFAR-100 dataset consists of 60000 32x32 colour images in 100 classes, with 600 images
per class. There are 500 training images and 100 testing images per class. There are 50000 training images and 10000 test images. The 100 classes are grouped into 20 superclasses.
There are two labels per image - fine label (actual class) and coarse label (superclass).
Supported Tasks and Leaderboards
image-classification: The… See the full description on the dataset page: https://huggingface.co/datasets/mteb/cifar100.xai-attack-detection-cifar10
XAI Attack Detection — CIFAR-10 PGD
This private research dataset contains balanced, paired clean and adversarial images for
studying whether an attack can be detected from a classifier explanation map.
Dataset construction
The source is the CIFAR-10 test split. A fine-tuned OpenCLIP ViT-B/16 classifies each
clean image. Clean-correct examples are attacked with untargeted L-infinity PGD using
epsilon 8/255, step size 2/255, 10 steps, and deterministic random… See the full description on the dataset page: https://huggingface.co/datasets/nimaeb/xai-attack-detection-cifar10.cifar10_quality_driftThis dataset was crafted to be used in our tutorial [Link to the tutorial when
ready]. It consists on product reviews from an e-commerce store. The reviews
are labeled on a scale from 1 to 5 (stars). The training & validation sets are
fully composed by reviews written in english. However, the production set has
some reviews written in spanish. At Arize, we work to surface this issue and
help you solve it.ciln-bench-cifar10
CILN-Bench: CIFAR-10
This is a label noise dataset. We corrupt the CIFAR-10 noisy-label-train split, 22,500 images, with a known corruption type and severity. Four trained classifiers, called voters, label the corrupted inputs. Each voter's top class is one vote. The vote shares are the label distribution of the input. The true labels, every voter's softmax and the corruption seeds are included, so you can build any kind of label you want.
Settings
15 corruption… See the full description on the dataset page: https://huggingface.co/datasets/sh-islam/ciln-bench-cifar10.cifar10h
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… See the full description on the dataset page: https://huggingface.co/datasets/MKZuziak/cifar10h.mlcd-mteb-cifar-eval
MLCD vs CLIP on MTEB CIFAR-10/100: integration and evaluation
Evaluation results accompanying the MTEB integration of two MLCD image encoders
(PR #5406, resolving
issue #2571).
Two DeepGlint-AI MLCD encoders were integrated into MTEB, verified against the
reference implementation, and evaluated on the official MTEB CIFAR-10/CIFAR-100
image-classification tasks alongside size-matched OpenAI CLIP baselines.
What was measured
Official MTEB image classification: 5… See the full description on the dataset page: https://huggingface.co/datasets/b4ph/mlcd-mteb-cifar-eval.latent-CIFAR100
Latent CIFAR100
This is the CIFAR100 dataset that has been latently encoded with various VAEs and saved as safetensors.
Paths are structured:
train/class/number.safetensors
test/class/number.safetensors
The files are in class named folders, each safetensor file contains its image data in the "latent" key, and the original class number in the "class" key (which I don't recommend using.)
I recommend using the sdxl-488 version! The 488 refers to the size of the latents (channels… See the full description on the dataset page: https://huggingface.co/datasets/Verah/latent-CIFAR100.cifar10
CIFAR-10 Dataset
Overview
CIFAR-10 is a widely used image classification dataset consisting of 60,000 color images of size 32×32 pixels. The dataset contains 10 classes with 6,000 images per class.
Classes
Airplane
Automobile
Bird
Cat
Deer
Dog
Frog
Horse
Ship
Truck
Dataset Structure & Statistics
cifar10/
-train/
-test/
Split
Images
Train
50,000
Test
10,000
Total
60,000
Original Source
CIFAR-10… See the full description on the dataset page: https://huggingface.co/datasets/ckyrkou/cifar10.cifar100-lt
Dataset Card for CIFAR-100-LT (Long Tail)
Dataset Summary
Note (March 2026): This dataset has been migrated from a Python loading script to
parquet format for compatibility with datasets v4.4+. No trust_remote_code=True
is needed. Available configs: r-10, r-20, r-50, r-100.
The CIFAR-100-LT imbalanced dataset is comprised of under 60,000 color images, each measuring 32x32 pixels,
distributed across 100 distinct classes.
The number of samples within each class… See the full description on the dataset page: https://huggingface.co/datasets/Amanmeena004/cifar100-lt.cifar10-lance
CIFAR-10 (Lance Format)
A Lance-formatted version of CIFAR-10 covering 60,000 32×32 RGB images across ten balanced object classes. Each row carries inline PNG bytes, the integer label, the human-readable class name, and a cosine-normalized CLIP image embedding, all backed by a bundled IVF_PQ vector index plus scalar indices on the label columns and available directly from the Hub at hf://datasets/lance-format/cifar10-lance/data.
Key features
Inline PNG bytes in the… See the full description on the dataset page: https://huggingface.co/datasets/lance-format/cifar10-lance.svae-freckles-4096-cifar10
SVAE Freckles 4096 — CIFAR-10 Omega Tokens
Precomputed spectral decomposition of CIFAR-10 through Freckles v41 (256×256), a frozen Spectral Variational Autoencoder trained exclusively on synthetic noise.
Each CIFAR-10 image is resized to 256×256, decomposed into 4096 patches (4×4 each), and passed through Freckles' encoder → SVD bottleneck. The 4 singular values per patch are stored as a (4, 64, 64) omega map — a 4-channel spatial representation of spectral energy.… See the full description on the dataset page: https://huggingface.co/datasets/AbstractPhil/svae-freckles-4096-cifar10.cifar10
Dataset Card for CIFAR-10
Dataset Summary
The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images per class. There are 50000 training images and 10000 test images.
The dataset is divided into five training batches and one test batch, each with 10000 images. The test batch contains exactly 1000 randomly-selected images from each class. The training batches contain the remaining images in random order, but some training batches may… See the full description on the dataset page: https://huggingface.co/datasets/puneet44/cifar10.cifar10-outlier
Dataset Card for "cifar10-outlier"
📚 This dataset is an enriched version of the CIFAR-10 Dataset.
The workflow is described in the medium article: Changes of Embeddings during Fine-Tuning of Transformers.
Explore the Dataset
The open source data curation tool Renumics Spotlight allows you to explorer this dataset. You can find a Hugging Face Spaces running Spotlight with this dataset here:
Full Version (High hardware requirement)… See the full description on the dataset page: https://huggingface.co/datasets/renumics/cifar10-outlier.CIFAR-10_Subset
CIFAR-10 — Subset
Stratified random subset of CIFAR-10.
Split
Rows
Per class
train
5,000
500
test
1,000
100
validation
500
50
Classes: airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck
Images: 32 × 32 RGB | Seed: 42
Label Map
ID
Class
ID
Class
0
airplane
5
dog
1
automobile
6
frog
2
bird
7
horse
3
cat
8
ship
4deer
9
truck
Usage
from datasets import load_dataset
ds =… See the full description on the dataset page: https://huggingface.co/datasets/Chiranjeev007/CIFAR-10_Subset.epinet_cifar10_logits
Cached CIFAR-10 test logits for the epinet tutorial
Test-set logits for six uncertainty-quantification agents on CIFAR-10, cached so
that the joint-prediction figures in the
lightning-uq-box epinet
tutorial can be rebuilt on CPU in seconds, without a GPU and without retraining
32 ResNets.
Consumed by docs/tutorials/classification/epinet.ipynb.
Contents
File
Description
test_logits.pt
torch.save dict of logit tensors (below)
metrics.csv
Scalar metrics… See the full description on the dataset page: https://huggingface.co/datasets/lightning-uq-box/epinet_cifar10_logits.