Openimage-O: Out-of-Distribution Detection Benchmark Derived from Open Images

Openimage-O is an out-of-distribution benchmark subset derived from the Open Images dataset. It originates from the OpenOOD benchmark evaluation splits. The dataset was uploaded by the user 'torch-uncertainty' and was last updated on October 10, 2025.

Use Cases

• Benchmarking out-of-distribution detection performance based on curated OOD splits.
• Training and evaluating model uncertainty based on a standardized OOD evaluation framework.
• Comparing OOD detection methods using a common subset of the large-scale Open Images dataset.

Strengths

• Derived from the large-scale Open Images dataset, which suggests a substantial base of image data.
• Specifically designed as a benchmark for out-of-distribution detection, providing a standardized evaluation target.

Limitations

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

Provenance

Source

Open Images dataset (Google)

Collection Method

Derived from the OpenOOD benchmark evaluation splits.

Freshness

Last updated 2025-10-10 13:44:07; freshness should be verified.