Benchmark RadarAI BENCHMARK PROFILE
OAO-AttackBench
OAO-AttackBench is a set of counterfactual prompts for one-and-only objects, used to evaluate text-to-image alignment in a specific study.
- Released
- 2026-06-29
- Readiness
- Inspectable
- Primary field
- General AI
Why it matters
The benchmark supports evaluating prompt adherence for concepts with strong visual priors, but it is introduced primarily to validate the proposed method and lacks a standalone public comparison path.
Motivation
Text-to-image (T2I) diffusion models often fail to faithfully render explicit textual descriptions, instead defaulting to strongly learned visual priors due to a phenomenon referred to as concept association bias.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.