TTA-ImageNet-C: ImageNet-C Benchmark for Test-Time Adaptation

TTA-ImageNet-C is a mirror of the ImageNet-C dataset created by Hendrycks & Dietterich for ICLR 2019. The dataset is maintained as part of the TTA-Evaluation-Harness and is intended for reproducible test-time adaptation evaluation. Its upstream source is Zenodo record 2235448.

Use Cases

• Benchmarking neural network robustness to common corruptions based on the dataset's stated purpose.
• Evaluating test-time adaptation algorithms based on the dataset's revision pin for reproducibility.
• Comparing model performance on perturbed images based on the dataset's nature as a corruption benchmark.

Strengths

• Dataset is derived from a seminal benchmark (ImageNet-C) published at ICLR 2019.
• Includes a revision pin for reproducible evaluation, as stated in the description.
• License is explicitly noted as CC BY 4.0, matching the upstream source.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file size are unknown, which may limit suitability assessment.
• The description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Zenodo record 2235448 (ImageNet-C by Hendrycks & Dietterich).

Collection Method

Mirror of a benchmark dataset for reproducible evaluation.

Time Range

null

Freshness

Last updated 2026-04-28 07:12:41; freshness should be verified.

Geography

null