Jingkang50/OpenOOD resource

Health v2 · maintenance only

Adoption not part of the score

1070 stars · 179 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

OpenOOD is a benchmark framework that reproduces and fairly compares methods for generalized out-of-distribution detection, spanning anomaly detection, novelty detection, open set recognition, and OOD detection. It provides standardized datasets, 35+ implemented methods, and a public leaderboard for evaluating detection performance.

Use cases

• benchmark out-of-distribution detection methods
• compare anomaly detection algorithms fairly
• evaluate open set recognition models
• reproduce OOD detection baselines
• find the best novelty detection method for my model
• test classifier robustness to out-of-distribution inputs

When to choose

• you need a standardized, reproducible comparison of OOD detection methods
• you are doing research on anomaly, novelty, or open set detection and want baselines
• you want access to a leaderboard of 35+ detection methods on common benchmarks

When to avoid

• you need a production-ready OOD detection service rather than a research benchmark
• you need multimodal or text-based OOD detection, which is still future work (v2.0)
• you just want a simple plug-and-play anomaly detector without benchmarking overhead

Facets

dataset · maturity active

benchmarking machine-learning computer-vision testing machine-learning computer-vision data-science python out-of-distribution-detection anomaly-detection open-set-recognition novelty-detection outlier-detection research-benchmark model-robustness algorithms gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="Jingkang50/OpenOOD")