Concepts
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Concepts dataset
0.717CLIP-I Score
14
121 concepts out-of-distribution suites (held-out)
95.5Detection Rate
14
167 held-out concepts (test)
83.1Stage 1 Input Fidelity (VLM)
4
17 concepts (evaluation set)
11Model Preference Count (Handoff)
2