Benchmark RadarWSADBench
WSADBench is a benchmark for weakly supervised anomaly detection (WSAD) that unifies evaluation across incomplete, inexact, and inaccurate supervision scenarios. It evaluates 36 algorithms across 4 modalities (tabular, video, image features, text embeddings) by systematically varying label quantity, granularity, and quality, with protocols and code provided.
- Released
- 2026-05-25
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
The field of weakly supervised anomaly detection has lacked a unified evaluation framework, with existing benchmarks isolating the three supervision types. WSADBench provides a standardized comparison that reveals performance boundaries across scenarios and informs algorithm selection for practitioners facing limited or noisy labels.
Motivation
Weakly supervised anomaly detection (WSAD) has developed in three primary directions: incomplete, inexact, and inaccurate supervision.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.