datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
stl10-c
STL-10-C
A corrupted variant of the STL-10 test split, generated for controlled mixed-batch OOD experiments.
Dataset Description
600,000 corrupted images produced by applying imagecorruptions to STL-10's 8,000-image test split across 15 corruption types and 5 severity levels.
Each record contains:
image: PNG bytes
label: integer class label (0–9)
corruption_name: corruption type (e.g. gaussian_noise)
corruption_level: severity (1–5)
Intended Use… See the full description on the dataset page: https://huggingface.co/datasets/Kipardox/stl10-c.STL-10_Subset
STL-10 — Subset
Stratified random subset of STL-10.
Split
Rows
Per class
train
5,000
500
test
1,000
100
validation
500
50
Classes: airplane, bird, car, cat, deer, dog, horse, monkey, ship, truck
Images: 96 × 96 RGB | Seed: 42
from datasets import load_dataset
ds = load_dataset("Chiranjeev007/STL-10_Subset")
sample = ds["train"][0]
sample["image"] # PIL Image 96×96 RGB
sample["label"] # int 0–9