Cross-modal Retrieval on AEA 553 events (test)
54.6Top-5 RecallSigLIP 2
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SigLIP 2Selection strategy=Novelty (ε-net)2026.03 | 54.6 | 0.003 | |
| SigLIP 2Selection strategy=Farthest-Point2026.03 | 52.8 | 0.003 | |
| SigLIP 2Selection strategy=K-Means2026.03 | 50.8 | 0.003 | |
| SigLIP 2Selection strategy=Uniform2026.03 | 50.5 | 0.003 | |
| ImageBindSelection strategy=Farthest-Point2026.03 | 45.4 | 0.007 | |
| SigLIP 2Selection strategy=Full2026.03 | 45.2 | 0.003 | |
| SigLIP 2Selection strategy=Random2026.03 | 44.6 | 0.003 | |
| ImageBindSelection strategy=Novelty (ε-net)2026.03 | 43.8 | 0.007 | |
| ImageBindSelection strategy=K-Means2026.03 | 43.4 | 0.007 | |
| CLIP L/14Selection strategy=Farthest-Point2026.03 | 41.6 | 0.12 | |
| CLIP L/14Selection strategy=Novelty (ε-net)2026.03 | 41.4 | 0.12 | |
| ImageBindSelection strategy=Uniform2026.03 | 41.2 | 0.007 | |
| CLIP L/14Selection strategy=K-Means2026.03 | 40.3 | 0.12 | |
| CLIP L/14Selection strategy=Uniform2026.03 | 40.3 | 0.12 | |
| CLIP B/32†Selection strategy=Novelty (ε-net)2026.03 | 40.3 | 0.001 | |
| SigLIPSelection strategy=Uniform2026.03 | 39.6 | 0.27 | |
| SigLIPSelection strategy=Novelty (ε-net)2026.03 | 39.2 | 0.27 | |
| CLIP B/32†Selection strategy=Farthest-Point2026.03 | 38.7 | 0.001 | |
| SigLIPSelection strategy=Farthest-Point2026.03 | 37.6 | 0.27 | |
| SigLIPSelection strategy=K-Means2026.03 | 37.4 | 0.27 | |
| ImageBindSelection strategy=Random2026.03 | 37.2 | 0.007 | |
| ImageBindSelection strategy=Full2026.03 | 36.7 | 0.007 | |
| CLIP L/14Selection strategy=Full2026.03 | 36.5 | 0.12 | |
| CLIP B/32†Selection strategy=K-Means2026.03 | 36.5 | 0.001 | |
| CLIP L/14Selection strategy=Random2026.03 | 35.9 | 0.12 | |
| MobileCLIPSelection strategy=Novelty (ε-net)2026.03 | 35.8 | 0.001 | |
| CLIP B/32†Selection strategy=Uniform2026.03 | 35.6 | 0.001 | |
| MobileCLIPSelection strategy=Farthest-Point2026.03 | 34.7 | 0.001 | |
| SigLIPSelection strategy=Full2026.03 | 34 | 0.27 | |
| MobileCLIPSelection strategy=K-Means2026.03 | 33.1 | 0.001 | |
| SigLIPSelection strategy=Random2026.03 | 32.8 | 0.27 | |
| MobileCLIPSelection strategy=Uniform2026.03 | 31.1 | 0.001 | |
| CLIP B/32†Selection strategy=Full2026.03 | 30.7 | 0.001 | |
| CLIP B/32†Selection strategy=Random2026.03 | 30.4 | 0.001 | |
| MobileCLIPSelection strategy=Random2026.03 | 28.3 | 0.001 | |
| MobileCLIPSelection strategy=Full2026.03 | 26.9 | 0.001 | |
| TinyCLIP+LiTSelection strategy=K-Means2026.03 | 22.1 | 0.002 | |
| TinyCLIP+LiTSelection strategy=Novelty (ε-net)2026.03 | 21.3 | 0.002 | |
| TinyCLIP+LiTSelection strategy=Uniform2026.03 | 21.2 | 0.002 | |
| TinyCLIP+LiTSelection strategy=Farthest-Point2026.03 | 18.4 | 0.002 | |
| TinyCLIP+LiTSelection strategy=Random2026.03 | 18 | 0.002 | |
| TinyCLIP+LiTSelection strategy=Full2026.03 | 15.2 | 0.002 | |
| TinyCLIPSelection strategy=Novelty (ε-net)2026.03 | 14.3 | 0.021 | |
| TinyCLIPSelection strategy=K-Means2026.03 | 13.9 | 0.021 | |
| TinyCLIPSelection strategy=Uniform2026.03 | 13.9 | 0.021 | |
| TinyCLIPSelection strategy=Farthest-Point2026.03 | 13.2 | 0.021 | |
| TinyCLIPSelection strategy=Random2026.03 | 12.8 | 0.021 | |
| TinyCLIPSelection strategy=Full2026.03 | 11 | 0.021 |