Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

CREPE

Benchmarks

Task NameDataset NameSOTA ResultTrend
Compositionality EvaluationCREPE
CU92.6
14
Image-to-Text retrievalCREPE (test)
Multi-Obj Acc0.8735
13
Image-to-text retrievalCREPE productivity
R@1 (replace)17.9
8
Image-to-text retrievalCREPE systematicity CC12M
R@1 (atom)36.6
8
Text-image retrievalCREPE
NDCG@1073.56
6
Compositional EvaluationCREPE productivity (test)
Replace Rate0.14
4
Compositional EvaluationCREPE systematicity (test)
Atom Score34
4
Showing 7 of 7 rows