Dataset Viewer
Auto-converted to Parquet Duplicate
The dataset viewer is not available for this split.
Parquet error: Scan size limit exceeded: attempted to read 555054087 bytes, limit is 300000000 bytes Make sure that 1. the Parquet files contain a page index to enable random access without loading entire row groups2. otherwise use smaller row-group sizes when serializing the Parquet files
Error code:   TooBigContentError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

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

Evaluate test-time adaptation methods (e.g. TENT) under controlled distribution shift, where the fraction of corrupted samples in a batch is the independent variable.

Source

STL-10 test split — Coates et al. (2011). Corruptions applied via imagecorruptions — Michaelis et al. (2019).

Downloads last month
52